LineageOS/android_external_libvpx
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vp9/encoder: apply clang-format

Browse Source 
Change-Id: I45d9fb4013f50766b24363a86365e8063e8954c2
This commit is contained in:
clang-format
2016-07-26 20:43:23 -07:00
committed by James Zern
parent 3a04c9c9c4
commit e0cc52db3f
81 changed files with 6382 additions and 8308 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
vp9/encoder/arm/neon/vp9_dct_neon.c
View File

@@ -18,14 +18,13 @@
#include "vpx_dsp/txfm_common.h"
void vp9_fdct8x8_quant_neon(const int16_t *input, int stride,
int16_t* coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t* zbin_ptr,
const int16_t* round_ptr, const int16_t* quant_ptr,
const int16_t* quant_shift_ptr,
int16_t* qcoeff_ptr, int16_t* dqcoeff_ptr,
const int16_t* dequant_ptr, uint16_t* eob_ptr,
const int16_t* scan_ptr,
const int16_t* iscan_ptr) {
int16_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr,
int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan_ptr,
const int16_t *iscan_ptr) {
int16_t temp_buffer[64];
(void)coeff_ptr;

33
vp9/encoder/arm/neon/vp9_quantize_neon.c
View File

@@ -26,8 +26,8 @@ void vp9_quantize_fp_neon(const int16_t *coeff_ptr, intptr_t count,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr,
int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
uint16_t *eob_ptr, const int16_t *scan,
const int16_t *iscan) {
// TODO(jingning) Decide the need of these arguments after the
// quantization process is completed.
(void)zbin_ptr;
@@ -54,12 +54,12 @@ void vp9_quantize_fp_neon(const int16_t *coeff_ptr, intptr_t count,
const int16x8_t v_coeff = vld1q_s16(&coeff_ptr[0]);
const int16x8_t v_coeff_sign = vshrq_n_s16(v_coeff, 15);
const int16x8_t v_tmp = vabaq_s16(v_round, v_coeff, v_zero);
const int32x4_t v_tmp_lo = vmull_s16(vget_low_s16(v_tmp),
vget_low_s16(v_quant));
const int32x4_t v_tmp_hi = vmull_s16(vget_high_s16(v_tmp),
vget_high_s16(v_quant));
const int16x8_t v_tmp2 = vcombine_s16(vshrn_n_s32(v_tmp_lo, 16),
vshrn_n_s32(v_tmp_hi, 16));
const int32x4_t v_tmp_lo =
vmull_s16(vget_low_s16(v_tmp), vget_low_s16(v_quant));
const int32x4_t v_tmp_hi =
vmull_s16(vget_high_s16(v_tmp), vget_high_s16(v_quant));
const int16x8_t v_tmp2 =
vcombine_s16(vshrn_n_s32(v_tmp_lo, 16), vshrn_n_s32(v_tmp_hi, 16));
const uint16x8_t v_nz_mask = vceqq_s16(v_tmp2, v_zero);
const int16x8_t v_iscan_plus1 = vaddq_s16(v_iscan, v_one);
const int16x8_t v_nz_iscan = vbslq_s16(v_nz_mask, v_zero, v_iscan_plus1);
@@ -79,12 +79,12 @@ void vp9_quantize_fp_neon(const int16_t *coeff_ptr, intptr_t count,
const int16x8_t v_coeff = vld1q_s16(&coeff_ptr[i]);
const int16x8_t v_coeff_sign = vshrq_n_s16(v_coeff, 15);
const int16x8_t v_tmp = vabaq_s16(v_round, v_coeff, v_zero);
const int32x4_t v_tmp_lo = vmull_s16(vget_low_s16(v_tmp),
vget_low_s16(v_quant));
const int32x4_t v_tmp_hi = vmull_s16(vget_high_s16(v_tmp),
vget_high_s16(v_quant));
const int16x8_t v_tmp2 = vcombine_s16(vshrn_n_s32(v_tmp_lo, 16),
vshrn_n_s32(v_tmp_hi, 16));
const int32x4_t v_tmp_lo =
vmull_s16(vget_low_s16(v_tmp), vget_low_s16(v_quant));
const int32x4_t v_tmp_hi =
vmull_s16(vget_high_s16(v_tmp), vget_high_s16(v_quant));
const int16x8_t v_tmp2 =
vcombine_s16(vshrn_n_s32(v_tmp_lo, 16), vshrn_n_s32(v_tmp_hi, 16));
const uint16x8_t v_nz_mask = vceqq_s16(v_tmp2, v_zero);
const int16x8_t v_iscan_plus1 = vaddq_s16(v_iscan, v_one);
const int16x8_t v_nz_iscan = vbslq_s16(v_nz_mask, v_zero, v_iscan_plus1);
@@ -96,9 +96,8 @@ void vp9_quantize_fp_neon(const int16_t *coeff_ptr, intptr_t count,
vst1q_s16(&dqcoeff_ptr[i], v_dqcoeff);
}
{
const int16x4_t v_eobmax_3210 =
vmax_s16(vget_low_s16(v_eobmax_76543210),
vget_high_s16(v_eobmax_76543210));
const int16x4_t v_eobmax_3210 = vmax_s16(
vget_low_s16(v_eobmax_76543210), vget_high_s16(v_eobmax_76543210));
const int64x1_t v_eobmax_xx32 =
vshr_n_s64(vreinterpret_s64_s16(v_eobmax_3210), 32);
const int16x4_t v_eobmax_tmp =

155
vp9/encoder/mips/msa/vp9_error_msa.c
View File

@@ -11,74 +11,73 @@
#include "./vp9_rtcd.h"
#include "vpx_dsp/mips/macros_msa.h"
#define BLOCK_ERROR_BLOCKSIZE_MSA(BSize) \
static int64_t block_error_##BSize##size_msa(const int16_t *coeff_ptr, \
const int16_t *dq_coeff_ptr, \
int64_t *ssz) { \
int64_t err = 0; \
uint32_t loop_cnt; \
v8i16 coeff, dq_coeff, coeff_r_h, coeff_l_h; \
v4i32 diff_r, diff_l, coeff_r_w, coeff_l_w; \
v2i64 sq_coeff_r, sq_coeff_l; \
v2i64 err0, err_dup0, err1, err_dup1; \
\
coeff = LD_SH(coeff_ptr); \
dq_coeff = LD_SH(dq_coeff_ptr); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DOTP_SW2_SD(coeff_r_w, coeff_l_w, coeff_r_w, coeff_l_w, \
sq_coeff_r, sq_coeff_l); \
DOTP_SW2_SD(diff_r, diff_l, diff_r, diff_l, err0, err1); \
\
coeff = LD_SH(coeff_ptr + 8); \
dq_coeff = LD_SH(dq_coeff_ptr + 8); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff_ptr += 16; \
dq_coeff_ptr += 16; \
\
for (loop_cnt = ((BSize >> 4) - 1); loop_cnt--;) { \
coeff = LD_SH(coeff_ptr); \
dq_coeff = LD_SH(dq_coeff_ptr); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff = LD_SH(coeff_ptr + 8); \
dq_coeff = LD_SH(dq_coeff_ptr + 8); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff_ptr += 16; \
dq_coeff_ptr += 16; \
} \
\
err_dup0 = __msa_splati_d(sq_coeff_r, 1); \
err_dup1 = __msa_splati_d(sq_coeff_l, 1); \
sq_coeff_r += err_dup0; \
sq_coeff_l += err_dup1; \
*ssz = __msa_copy_s_d(sq_coeff_r, 0); \
*ssz += __msa_copy_s_d(sq_coeff_l, 0); \
\
err_dup0 = __msa_splati_d(err0, 1); \
err_dup1 = __msa_splati_d(err1, 1); \
err0 += err_dup0; \
err1 += err_dup1; \
err = __msa_copy_s_d(err0, 0); \
err += __msa_copy_s_d(err1, 0); \
\
return err; \
}
#define BLOCK_ERROR_BLOCKSIZE_MSA(BSize) \
static int64_t block_error_##BSize##size_msa( \
const int16_t *coeff_ptr, const int16_t *dq_coeff_ptr, int64_t *ssz) { \
int64_t err = 0; \
uint32_t loop_cnt; \
v8i16 coeff, dq_coeff, coeff_r_h, coeff_l_h; \
v4i32 diff_r, diff_l, coeff_r_w, coeff_l_w; \
v2i64 sq_coeff_r, sq_coeff_l; \
v2i64 err0, err_dup0, err1, err_dup1; \
\
coeff = LD_SH(coeff_ptr); \
dq_coeff = LD_SH(dq_coeff_ptr); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DOTP_SW2_SD(coeff_r_w, coeff_l_w, coeff_r_w, coeff_l_w, sq_coeff_r, \
sq_coeff_l); \
DOTP_SW2_SD(diff_r, diff_l, diff_r, diff_l, err0, err1); \
\
coeff = LD_SH(coeff_ptr + 8); \
dq_coeff = LD_SH(dq_coeff_ptr + 8); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff_ptr += 16; \
dq_coeff_ptr += 16; \
\
for (loop_cnt = ((BSize >> 4) - 1); loop_cnt--;) { \
coeff = LD_SH(coeff_ptr); \
dq_coeff = LD_SH(dq_coeff_ptr); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff = LD_SH(coeff_ptr + 8); \
dq_coeff = LD_SH(dq_coeff_ptr + 8); \
UNPCK_SH_SW(coeff, coeff_r_w, coeff_l_w); \
ILVRL_H2_SH(coeff, dq_coeff, coeff_r_h, coeff_l_h); \
HSUB_UH2_SW(coeff_r_h, coeff_l_h, diff_r, diff_l); \
DPADD_SD2_SD(coeff_r_w, coeff_l_w, sq_coeff_r, sq_coeff_l); \
DPADD_SD2_SD(diff_r, diff_l, err0, err1); \
\
coeff_ptr += 16; \
dq_coeff_ptr += 16; \
} \
\
err_dup0 = __msa_splati_d(sq_coeff_r, 1); \
err_dup1 = __msa_splati_d(sq_coeff_l, 1); \
sq_coeff_r += err_dup0; \
sq_coeff_l += err_dup1; \
*ssz = __msa_copy_s_d(sq_coeff_r, 0); \
*ssz += __msa_copy_s_d(sq_coeff_l, 0); \
\
err_dup0 = __msa_splati_d(err0, 1); \
err_dup1 = __msa_splati_d(err1, 1); \
err0 += err_dup0; \
err1 += err_dup1; \
err = __msa_copy_s_d(err0, 0); \
err += __msa_copy_s_d(err1, 0); \
\
return err; \
}
BLOCK_ERROR_BLOCKSIZE_MSA(16);
BLOCK_ERROR_BLOCKSIZE_MSA(64);
@@ -86,25 +85,17 @@ BLOCK_ERROR_BLOCKSIZE_MSA(256);
BLOCK_ERROR_BLOCKSIZE_MSA(1024);
int64_t vp9_block_error_msa(const tran_low_t *coeff_ptr,
const tran_low_t *dq_coeff_ptr,
intptr_t blk_size, int64_t *ssz) {
const tran_low_t *dq_coeff_ptr, intptr_t blk_size,
int64_t *ssz) {
int64_t err;
const int16_t *coeff = (const int16_t *)coeff_ptr;
const int16_t *dq_coeff = (const int16_t *)dq_coeff_ptr;
switch (blk_size) {
case 16:
err = block_error_16size_msa(coeff, dq_coeff, ssz);
break;
case 64:
err = block_error_64size_msa(coeff, dq_coeff, ssz);
break;
case 256:
err = block_error_256size_msa(coeff, dq_coeff, ssz);
break;
case 1024:
err = block_error_1024size_msa(coeff, dq_coeff, ssz);
break;
case 16: err = block_error_16size_msa(coeff, dq_coeff, ssz); break;
case 64: err = block_error_64size_msa(coeff, dq_coeff, ssz); break;
case 256: err = block_error_256size_msa(coeff, dq_coeff, ssz); break;
case 1024: err = block_error_1024size_msa(coeff, dq_coeff, ssz); break;
default:
err = vp9_block_error_c(coeff_ptr, dq_coeff_ptr, blk_size, ssz);
break;

133
vp9/encoder/mips/msa/vp9_fdct16x16_msa.c
View File

@@ -159,8 +159,8 @@ static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
/* load input data */
LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
FDCT_POSTPROC_2V_NEG_H(r0, r1);
FDCT_POSTPROC_2V_NEG_H(r2, r3);
FDCT_POSTPROC_2V_NEG_H(r4, r5);
@@ -169,8 +169,8 @@ static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
out += 64;
LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
FDCT_POSTPROC_2V_NEG_H(r8, r9);
FDCT_POSTPROC_2V_NEG_H(r10, r11);
FDCT_POSTPROC_2V_NEG_H(r12, r13);
@@ -181,8 +181,8 @@ static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
/* load input data */
input += 128;
LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
FDCT_POSTPROC_2V_NEG_H(r0, r1);
FDCT_POSTPROC_2V_NEG_H(r2, r3);
FDCT_POSTPROC_2V_NEG_H(r4, r5);
@@ -191,8 +191,8 @@ static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
out += 64;
LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
FDCT_POSTPROC_2V_NEG_H(r8, r9);
FDCT_POSTPROC_2V_NEG_H(r10, r11);
FDCT_POSTPROC_2V_NEG_H(r12, r13);
@@ -339,24 +339,24 @@ static void fadst16_transpose_msa(int16_t *input, int16_t *out) {
v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
/* load input data */
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11,
l4, l12, l5, l13, l6, l14, l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
out += 16 * 8;
/* load input data */
input += 128;
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11,
l4, l12, l5, l13, l6, l14, l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
r0, r1, r2, r3, r4, r5, r6, r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
r8, r9, r10, r11, r12, r13, r14, r15);
LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11, l4, l12, l5, l13, l6, l14,
l7, l15);
TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7, r0, r1, r2, r3, r4, r5, r6,
r7);
TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15, r8, r9, r10, r11,
r12, r13, r14, r15);
ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
}
@@ -371,10 +371,10 @@ static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) {
LD_SH8(temp, 16, in0, in1, in2, in3, in4, in5, in6, in7);
temp = intermediate + 8;
LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
in10, in11, in12, in13, in14, in15);
FDCT_POSTPROC_2V_NEG_H(in0, in1);
FDCT_POSTPROC_2V_NEG_H(in2, in3);
FDCT_POSTPROC_2V_NEG_H(in4, in5);
@@ -383,29 +383,28 @@ static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) {
FDCT_POSTPROC_2V_NEG_H(in10, in11);
FDCT_POSTPROC_2V_NEG_H(in12, in13);
FDCT_POSTPROC_2V_NEG_H(in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
in8, in9, in10, in11, in12, in13, in14, in15);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6,
tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
temp = intermediate;
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, temp, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp0, tmp1,
tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
temp = intermediate;
LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3,
tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15, in0, in1, in2, in3,
in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, tmp0, in0,
tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, out, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7,
tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, tmp4, in4,
tmp5, in5, tmp6, in6, tmp7, in7);
out = output + 8;
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, out, 16);
}
void vp9_fht16x16_msa(const int16_t *input, int16_t *output,
int32_t stride, int32_t tx_type) {
void vp9_fht16x16_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
DECLARE_ALIGNED(32, int16_t, tmp[256]);
DECLARE_ALIGNED(32, int16_t, trans_buf[256]);
DECLARE_ALIGNED(32, int16_t, tmp_buf[128]);
@@ -413,35 +412,31 @@ void vp9_fht16x16_msa(const int16_t *input, int16_t *output,
int16_t *ptmpbuf = &tmp_buf[0];
int16_t *trans = &trans_buf[0];
const int32_t const_arr[29 * 4] = {
52707308, 52707308, 52707308, 52707308,
-1072430300, -1072430300, -1072430300, -1072430300,
795618043, 795618043, 795618043, 795618043,
-721080468, -721080468, -721080468, -721080468,
459094491, 459094491, 459094491, 459094491,
-970646691, -970646691, -970646691, -970646691,
1010963856, 1010963856, 1010963856, 1010963856,
-361743294, -361743294, -361743294, -361743294,
209469125, 209469125, 209469125, 209469125,
-1053094788, -1053094788, -1053094788, -1053094788,
1053160324, 1053160324, 1053160324, 1053160324,
639644520, 639644520, 639644520, 639644520,
-862444000, -862444000, -862444000, -862444000,
1062144356, 1062144356, 1062144356, 1062144356,
-157532337, -157532337, -157532337, -157532337,
260914709, 260914709, 260914709, 260914709,
-1041559667, -1041559667, -1041559667, -1041559667,
920985831, 920985831, 920985831, 920985831,
-551995675, -551995675, -551995675, -551995675,
596522295, 596522295, 596522295, 596522295,
892853362, 892853362, 892853362, 892853362,
-892787826, -892787826, -892787826, -892787826,
410925857, 410925857, 410925857, 410925857,
-992012162, -992012162, -992012162, -992012162,
992077698, 992077698, 992077698, 992077698,
759246145, 759246145, 759246145, 759246145,
-759180609, -759180609, -759180609, -759180609,
-759222975, -759222975, -759222975, -759222975,
759288511, 759288511, 759288511, 759288511 };
52707308, 52707308, 52707308, 52707308, -1072430300,
-1072430300, -1072430300, -1072430300, 795618043, 795618043,
795618043, 795618043, -721080468, -721080468, -721080468,
-721080468, 459094491, 459094491, 459094491, 459094491,
-970646691, -970646691, -970646691, -970646691, 1010963856,
1010963856, 1010963856, 1010963856, -361743294, -361743294,
-361743294, -361743294, 209469125, 209469125, 209469125,
209469125, -1053094788, -1053094788, -1053094788, -1053094788,
1053160324, 1053160324, 1053160324, 1053160324, 639644520,
639644520, 639644520, 639644520, -862444000, -862444000,
-862444000, -862444000, 1062144356, 1062144356, 1062144356,
1062144356, -157532337, -157532337, -157532337, -157532337,
260914709, 260914709, 260914709, 260914709, -1041559667,
-1041559667, -1041559667, -1041559667, 920985831, 920985831,
920985831, 920985831, -551995675, -551995675, -551995675,
-551995675, 596522295, 596522295, 596522295, 596522295,
892853362, 892853362, 892853362, 892853362, -892787826,
-892787826, -892787826, -892787826, 410925857, 410925857,
410925857, 410925857, -992012162, -992012162, -992012162,
-992012162, 992077698, 992077698, 992077698, 992077698,
759246145, 759246145, 759246145, 759246145, -759180609,
-759180609, -759180609, -759180609, -759222975, -759222975,
-759222975, -759222975, 759288511, 759288511, 759288511,
759288511
};
switch (tx_type) {
case DCT_DCT:
@@ -500,8 +495,6 @@ void vp9_fht16x16_msa(const int16_t *input, int16_t *output,
fadst16_transpose_msa(tmp, output);
break;
default:
assert(0);
break;
default: assert(0); break;
}
}

4
vp9/encoder/mips/msa/vp9_fdct4x4_msa.c
View File

@@ -86,9 +86,7 @@ void vp9_fht4x4_msa(const int16_t *input, int16_t *output, int32_t stride,
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FADST4(in0, in1, in2, in3, in0, in1, in2, in3);
break;
default:
assert(0);
break;
default: assert(0); break;
}
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);

56
vp9/encoder/mips/msa/vp9_fdct8x8_msa.c
View File

@@ -23,44 +23,42 @@ void vp9_fht8x8_msa(const int16_t *input, int16_t *output, int32_t stride,
switch (tx_type) {
case DCT_DCT:
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
break;
case ADST_DCT:
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
break;
case DCT_ADST:
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
break;
case ADST_ADST:
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
default:
assert(0);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2,
in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4,
in5, in6, in7);
break;
default: assert(0); break;
}
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
in4, in5, in6, in7);
SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
}

197
vp9/encoder/mips/msa/vp9_fdct_msa.h
View File

@@ -15,103 +15,102 @@
#include "vpx_dsp/mips/txfm_macros_msa.h"
#include "vpx_ports/mem.h"
#define VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 cnst0_m, cnst1_m, cnst2_m, cnst3_m, cnst4_m; \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m, s0_m, s1_m; \
v8i16 coeff0_m = { cospi_2_64, cospi_6_64, cospi_10_64, cospi_14_64, \
cospi_18_64, cospi_22_64, cospi_26_64, cospi_30_64 }; \
v8i16 coeff1_m = { cospi_8_64, -cospi_8_64, cospi_16_64, -cospi_16_64, \
cospi_24_64, -cospi_24_64, 0, 0 }; \
\
SPLATI_H2_SH(coeff0_m, 0, 7, cnst0_m, cnst1_m); \
cnst2_m = -cnst0_m; \
ILVEV_H2_SH(cnst0_m, cnst1_m, cnst1_m, cnst2_m, cnst0_m, cnst1_m); \
SPLATI_H2_SH(coeff0_m, 4, 3, cnst2_m, cnst3_m); \
cnst4_m = -cnst2_m; \
ILVEV_H2_SH(cnst2_m, cnst3_m, cnst3_m, cnst4_m, cnst2_m, cnst3_m); \
\
ILVRL_H2_SH(in0, in7, vec1_m, vec0_m); \
ILVRL_H2_SH(in4, in3, vec3_m, vec2_m); \
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, \
cnst1_m, cnst2_m, cnst3_m, in7, in0, \
in4, in3); \
\
SPLATI_H2_SH(coeff0_m, 2, 5, cnst0_m, cnst1_m); \
cnst2_m = -cnst0_m; \
ILVEV_H2_SH(cnst0_m, cnst1_m, cnst1_m, cnst2_m, cnst0_m, cnst1_m); \
SPLATI_H2_SH(coeff0_m, 6, 1, cnst2_m, cnst3_m); \
cnst4_m = -cnst2_m; \
ILVEV_H2_SH(cnst2_m, cnst3_m, cnst3_m, cnst4_m, cnst2_m, cnst3_m); \
\
ILVRL_H2_SH(in2, in5, vec1_m, vec0_m); \
ILVRL_H2_SH(in6, in1, vec3_m, vec2_m); \
\
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, \
cnst1_m, cnst2_m, cnst3_m, in5, in2, \
in6, in1); \
BUTTERFLY_4(in7, in0, in2, in5, s1_m, s0_m, in2, in5); \
out7 = -s0_m; \
out0 = s1_m; \
\
SPLATI_H4_SH(coeff1_m, 0, 4, 1, 5, cnst0_m, cnst1_m, cnst2_m, cnst3_m); \
\
ILVEV_H2_SH(cnst3_m, cnst0_m, cnst1_m, cnst2_m, cnst3_m, cnst2_m); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
cnst1_m = cnst0_m; \
\
ILVRL_H2_SH(in4, in3, vec1_m, vec0_m); \
ILVRL_H2_SH(in6, in1, vec3_m, vec2_m); \
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, \
cnst2_m, cnst3_m, cnst1_m, out1, out6, \
s0_m, s1_m); \
\
SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
ILVRL_H2_SH(in2, in5, vec1_m, vec0_m); \
ILVRL_H2_SH(s0_m, s1_m, vec3_m, vec2_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m); \
\
out1 = -out1; \
out3 = -out3; \
out5 = -out5; \
}
#define VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, \
out3, out4, out5, out6, out7) \
{ \
v8i16 cnst0_m, cnst1_m, cnst2_m, cnst3_m, cnst4_m; \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m, s0_m, s1_m; \
v8i16 coeff0_m = { cospi_2_64, cospi_6_64, cospi_10_64, cospi_14_64, \
cospi_18_64, cospi_22_64, cospi_26_64, cospi_30_64 }; \
v8i16 coeff1_m = { cospi_8_64, -cospi_8_64, cospi_16_64, -cospi_16_64, \
cospi_24_64, -cospi_24_64, 0, 0 }; \
\
SPLATI_H2_SH(coeff0_m, 0, 7, cnst0_m, cnst1_m); \
cnst2_m = -cnst0_m; \
ILVEV_H2_SH(cnst0_m, cnst1_m, cnst1_m, cnst2_m, cnst0_m, cnst1_m); \
SPLATI_H2_SH(coeff0_m, 4, 3, cnst2_m, cnst3_m); \
cnst4_m = -cnst2_m; \
ILVEV_H2_SH(cnst2_m, cnst3_m, cnst3_m, cnst4_m, cnst2_m, cnst3_m); \
\
ILVRL_H2_SH(in0, in7, vec1_m, vec0_m); \
ILVRL_H2_SH(in4, in3, vec3_m, vec2_m); \
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, cnst1_m, \
cnst2_m, cnst3_m, in7, in0, in4, in3); \
\
SPLATI_H2_SH(coeff0_m, 2, 5, cnst0_m, cnst1_m); \
cnst2_m = -cnst0_m; \
ILVEV_H2_SH(cnst0_m, cnst1_m, cnst1_m, cnst2_m, cnst0_m, cnst1_m); \
SPLATI_H2_SH(coeff0_m, 6, 1, cnst2_m, cnst3_m); \
cnst4_m = -cnst2_m; \
ILVEV_H2_SH(cnst2_m, cnst3_m, cnst3_m, cnst4_m, cnst2_m, cnst3_m); \
\
ILVRL_H2_SH(in2, in5, vec1_m, vec0_m); \
ILVRL_H2_SH(in6, in1, vec3_m, vec2_m); \
\
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, cnst1_m, \
cnst2_m, cnst3_m, in5, in2, in6, in1); \
BUTTERFLY_4(in7, in0, in2, in5, s1_m, s0_m, in2, in5); \
out7 = -s0_m; \
out0 = s1_m; \
\
SPLATI_H4_SH(coeff1_m, 0, 4, 1, 5, cnst0_m, cnst1_m, cnst2_m, cnst3_m); \
\
ILVEV_H2_SH(cnst3_m, cnst0_m, cnst1_m, cnst2_m, cnst3_m, cnst2_m); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
cnst1_m = cnst0_m; \
\
ILVRL_H2_SH(in4, in3, vec1_m, vec0_m); \
ILVRL_H2_SH(in6, in1, vec3_m, vec2_m); \
DOT_ADD_SUB_SRARI_PCK(vec0_m, vec1_m, vec2_m, vec3_m, cnst0_m, cnst2_m, \
cnst3_m, cnst1_m, out1, out6, s0_m, s1_m); \
\
SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
ILVRL_H2_SH(in2, in5, vec1_m, vec0_m); \
ILVRL_H2_SH(s0_m, s1_m, vec3_m, vec2_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m); \
\
out1 = -out1; \
out3 = -out3; \
out5 = -out5; \
}
#define VP9_FADST4(in0, in1, in2, in3, out0, out1, out2, out3) { \
v4i32 s0_m, s1_m, s2_m, s3_m, constant_m; \
v4i32 in0_r_m, in1_r_m, in2_r_m, in3_r_m; \
\
UNPCK_R_SH_SW(in0, in0_r_m); \
UNPCK_R_SH_SW(in1, in1_r_m); \
UNPCK_R_SH_SW(in2, in2_r_m); \
UNPCK_R_SH_SW(in3, in3_r_m); \
\
constant_m = __msa_fill_w(sinpi_4_9); \
MUL2(in0_r_m, constant_m, in3_r_m, constant_m, s1_m, s0_m); \
\
constant_m = __msa_fill_w(sinpi_1_9); \
s0_m += in0_r_m * constant_m; \
s1_m -= in1_r_m * constant_m; \
\
constant_m = __msa_fill_w(sinpi_2_9); \
s0_m += in1_r_m * constant_m; \
s1_m += in3_r_m * constant_m; \
\
s2_m = in0_r_m + in1_r_m - in3_r_m; \
\
constant_m = __msa_fill_w(sinpi_3_9); \
MUL2(in2_r_m, constant_m, s2_m, constant_m, s3_m, in1_r_m); \
\
in0_r_m = s0_m + s3_m; \
s2_m = s1_m - s3_m; \
s3_m = s1_m - s0_m + s3_m; \
\
SRARI_W4_SW(in0_r_m, in1_r_m, s2_m, s3_m, DCT_CONST_BITS); \
PCKEV_H4_SH(in0_r_m, in0_r_m, in1_r_m, in1_r_m, s2_m, s2_m, \
s3_m, s3_m, out0, out1, out2, out3); \
}
#endif /* VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_ */
#define VP9_FADST4(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
v4i32 s0_m, s1_m, s2_m, s3_m, constant_m; \
v4i32 in0_r_m, in1_r_m, in2_r_m, in3_r_m; \
\
UNPCK_R_SH_SW(in0, in0_r_m); \
UNPCK_R_SH_SW(in1, in1_r_m); \
UNPCK_R_SH_SW(in2, in2_r_m); \
UNPCK_R_SH_SW(in3, in3_r_m); \
\
constant_m = __msa_fill_w(sinpi_4_9); \
MUL2(in0_r_m, constant_m, in3_r_m, constant_m, s1_m, s0_m); \
\
constant_m = __msa_fill_w(sinpi_1_9); \
s0_m += in0_r_m * constant_m; \
s1_m -= in1_r_m * constant_m; \
\
constant_m = __msa_fill_w(sinpi_2_9); \
s0_m += in1_r_m * constant_m; \
s1_m += in3_r_m * constant_m; \
\
s2_m = in0_r_m + in1_r_m - in3_r_m; \
\
constant_m = __msa_fill_w(sinpi_3_9); \
MUL2(in2_r_m, constant_m, s2_m, constant_m, s3_m, in1_r_m); \
\
in0_r_m = s0_m + s3_m; \
s2_m = s1_m - s3_m; \
s3_m = s1_m - s0_m + s3_m; \
\
SRARI_W4_SW(in0_r_m, in1_r_m, s2_m, s3_m, DCT_CONST_BITS); \
PCKEV_H4_SH(in0_r_m, in0_r_m, in1_r_m, in1_r_m, s2_m, s2_m, s3_m, s3_m, \
out0, out1, out2, out3); \
}
#endif /* VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_ */

78
vp9/encoder/mips/msa/vp9_temporal_filter_msa.c
View File

@@ -11,12 +11,9 @@
#include "./vp9_rtcd.h"
#include "vpx_dsp/mips/macros_msa.h"
static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
uint32_t stride,
uint8_t *frm2_ptr,
int32_t filt_sth,
int32_t filt_wgt,
uint32_t *acc,
static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr, uint32_t stride,
uint8_t *frm2_ptr, int32_t filt_sth,
int32_t filt_wgt, uint32_t *acc,
uint16_t *cnt) {
uint32_t row;
uint64_t f0, f1, f2, f3;
@@ -54,10 +51,10 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
HSUB_UB2_SH(frm_r, frm_l, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l, diff1_l,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3, mod0_w,
mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
@@ -65,8 +62,8 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w, mod0_w,
mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
@@ -85,8 +82,8 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3, mod0_w, mod1_w,
mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
@@ -101,10 +98,10 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
HSUB_UB2_SH(frm_r, frm_l, diff0, diff1);
UNPCK_SH_SW(diff0, diff0_r, diff0_l);
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l,
diff1_l, mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l, diff1_l,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3, mod0_w,
mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
@@ -112,8 +109,8 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w, mod0_w,
mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
@@ -131,8 +128,8 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3, mod0_w, mod1_w,
mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
@@ -141,13 +138,10 @@ static void temporal_filter_apply_8size_msa(uint8_t *frm1_ptr,
}
}
static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
uint32_t stride,
static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr, uint32_t stride,
uint8_t *frm2_ptr,
int32_t filt_sth,
int32_t filt_wgt,
uint32_t *acc,
uint16_t *cnt) {
int32_t filt_sth, int32_t filt_wgt,
uint32_t *acc, uint16_t *cnt) {
uint32_t row;
v16i8 frm1, frm2, frm3, frm4;
v16u8 frm_r, frm_l;
@@ -183,8 +177,8 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l, diff1_l,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3, mod0_w,
mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
@@ -192,8 +186,8 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w, mod0_w,
mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
@@ -212,8 +206,8 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3, mod0_w, mod1_w,
mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
@@ -230,8 +224,8 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
UNPCK_SH_SW(diff1, diff1_r, diff1_l);
MUL4(diff0_r, diff0_r, diff0_l, diff0_l, diff1_r, diff1_r, diff1_l, diff1_l,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3,
mod0_w, mod1_w, mod2_w, mod3_w);
MUL4(mod0_w, cnst3, mod1_w, cnst3, mod2_w, cnst3, mod3_w, cnst3, mod0_w,
mod1_w, mod2_w, mod3_w);
SRAR_W4_SW(mod0_w, mod1_w, mod2_w, mod3_w, strength);
diff0_r = (mod0_w < cnst16);
@@ -239,8 +233,8 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
diff1_r = (mod2_w < cnst16);
diff1_l = (mod3_w < cnst16);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w,
mod0_w, mod1_w, mod2_w, mod3_w);
SUB4(cnst16, mod0_w, cnst16, mod1_w, cnst16, mod2_w, cnst16, mod3_w, mod0_w,
mod1_w, mod2_w, mod3_w);
mod0_w = diff0_r & mod0_w;
mod1_w = diff0_l & mod1_w;
@@ -259,8 +253,8 @@ static void temporal_filter_apply_16size_msa(uint8_t *frm1_ptr,
UNPCK_SH_SW(frm2_l, frm2_lr, frm2_ll);
MUL4(mod0_w, frm2_rr, mod1_w, frm2_rl, mod2_w, frm2_lr, mod3_w, frm2_ll,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3,
mod0_w, mod1_w, mod2_w, mod3_w);
ADD4(mod0_w, acc0, mod1_w, acc1, mod2_w, acc2, mod3_w, acc3, mod0_w, mod1_w,
mod2_w, mod3_w);
ST_SW2(mod0_w, mod1_w, acc, 4);
acc += 8;
ST_SW2(mod2_w, mod3_w, acc, 4);
@@ -277,11 +271,11 @@ void vp9_temporal_filter_apply_msa(uint8_t *frame1_ptr, uint32_t stride,
int32_t filt_wgt, uint32_t *accu,
uint16_t *cnt) {
if (8 == (blk_w * blk_h)) {
temporal_filter_apply_8size_msa(frame1_ptr, stride, frame2_ptr,
strength, filt_wgt, accu, cnt);
temporal_filter_apply_8size_msa(frame1_ptr, stride, frame2_ptr, strength,
filt_wgt, accu, cnt);
} else if (16 == (blk_w * blk_h)) {
temporal_filter_apply_16size_msa(frame1_ptr, stride, frame2_ptr,
strength, filt_wgt, accu, cnt);
temporal_filter_apply_16size_msa(frame1_ptr, stride, frame2_ptr, strength,
filt_wgt, accu, cnt);
} else {
vp9_temporal_filter_apply_c(frame1_ptr, stride, frame2_ptr, blk_w, blk_h,
strength, filt_wgt, accu, cnt);

4
vp9/encoder/vp9_aq_360.c
View File

@@ -22,8 +22,8 @@
#include "vp9/encoder/vp9_rd.h"
#include "vp9/encoder/vp9_segmentation.h"
static const double rate_ratio[MAX_SEGMENTS] =
{1.0, 0.75, 0.6, 0.5, 0.4, 0.3, 0.25};
static const double rate_ratio[MAX_SEGMENTS] = { 1.0, 0.75, 0.6, 0.5,
0.4, 0.3, 0.25 };
// Sets segment id 0 for the equatorial region, 1 for temperate region
// and 2 for the polar regions

1
vp9/encoder/vp9_aq_360.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_AQ_360_H_
#define VP9_ENCODER_VP9_AQ_360_H_

57
vp9/encoder/vp9_aq_complexity.c
View File

@@ -19,21 +19,24 @@
#include "vp9/common/vp9_seg_common.h"
#include "vp9/encoder/vp9_segmentation.h"
#define AQ_C_SEGMENTS 5
#define DEFAULT_AQ2_SEG 3 // Neutral Q segment
#define AQ_C_SEGMENTS 5
#define DEFAULT_AQ2_SEG 3 // Neutral Q segment
#define AQ_C_STRENGTHS 3
static const double aq_c_q_adj_factor[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
{ {1.75, 1.25, 1.05, 1.00, 0.90},
{2.00, 1.50, 1.15, 1.00, 0.85},
{2.50, 1.75, 1.25, 1.00, 0.80} };
static const double aq_c_transitions[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
{ {0.15, 0.30, 0.55, 2.00, 100.0},
{0.20, 0.40, 0.65, 2.00, 100.0},
{0.25, 0.50, 0.75, 2.00, 100.0} };
static const double aq_c_var_thresholds[AQ_C_STRENGTHS][AQ_C_SEGMENTS] =
{ {-4.0, -3.0, -2.0, 100.00, 100.0},
{-3.5, -2.5, -1.5, 100.00, 100.0},
{-3.0, -2.0, -1.0, 100.00, 100.0} };
static const double aq_c_q_adj_factor[AQ_C_STRENGTHS][AQ_C_SEGMENTS] = {
{ 1.75, 1.25, 1.05, 1.00, 0.90 },
{ 2.00, 1.50, 1.15, 1.00, 0.85 },
{ 2.50, 1.75, 1.25, 1.00, 0.80 }
};
static const double aq_c_transitions[AQ_C_STRENGTHS][AQ_C_SEGMENTS] = {
{ 0.15, 0.30, 0.55, 2.00, 100.0 },
{ 0.20, 0.40, 0.65, 2.00, 100.0 },
{ 0.25, 0.50, 0.75, 2.00, 100.0 }
};
static const double aq_c_var_thresholds[AQ_C_STRENGTHS][AQ_C_SEGMENTS] = {
{ -4.0, -3.0, -2.0, 100.00, 100.0 },
{ -3.5, -2.5, -1.5, 100.00, 100.0 },
{ -3.0, -2.0, -1.0, 100.00, 100.0 }
};
static int get_aq_c_strength(int q_index, vpx_bit_depth_t bit_depth) {
// Approximate base quatizer (truncated to int)
@@ -78,14 +81,11 @@ void vp9_setup_in_frame_q_adj(VP9_COMP *cpi) {
for (segment = 0; segment < AQ_C_SEGMENTS; ++segment) {
int qindex_delta;
if (segment == DEFAULT_AQ2_SEG)
continue;
qindex_delta =
vp9_compute_qdelta_by_rate(&cpi->rc, cm->frame_type, cm->base_qindex,
aq_c_q_adj_factor[aq_strength][segment],
cm->bit_depth);
if (segment == DEFAULT_AQ2_SEG) continue;
qindex_delta = vp9_compute_qdelta_by_rate(
&cpi->rc, cm->frame_type, cm->base_qindex,
aq_c_q_adj_factor[aq_strength][segment], cm->bit_depth);
// For AQ complexity mode, we dont allow Q0 in a segment if the base
// Q is not 0. Q0 (lossless) implies 4x4 only and in AQ mode 2 a segment
@@ -125,26 +125,25 @@ void vp9_caq_select_segment(VP9_COMP *cpi, MACROBLOCK *mb, BLOCK_SIZE bs,
} else {
// Rate depends on fraction of a SB64 in frame (xmis * ymis / bw * bh).
// It is converted to bits * 256 units.
const int target_rate = (cpi->rc.sb64_target_rate * xmis * ymis * 256) /
(bw * bh);
const int target_rate =
(cpi->rc.sb64_target_rate * xmis * ymis * 256) / (bw * bh);
double logvar;
double low_var_thresh;
const int aq_strength = get_aq_c_strength(cm->base_qindex, cm->bit_depth);
vpx_clear_system_state();
low_var_thresh = (cpi->oxcf.pass == 2)
? VPXMAX(cpi->twopass.mb_av_energy, MIN_DEFAULT_LV_THRESH)
: DEFAULT_LV_THRESH;
low_var_thresh = (cpi->oxcf.pass == 2) ? VPXMAX(cpi->twopass.mb_av_energy,
MIN_DEFAULT_LV_THRESH)
: DEFAULT_LV_THRESH;
vp9_setup_src_planes(mb, cpi->Source, mi_row, mi_col);
logvar = vp9_log_block_var(cpi, mb, bs);
segment = AQ_C_SEGMENTS - 1; // Just in case no break out below.
segment = AQ_C_SEGMENTS - 1; // Just in case no break out below.
for (i = 0; i < AQ_C_SEGMENTS; ++i) {
// Test rate against a threshold value and variance against a threshold.
// Increasing segment number (higher variance and complexity) = higher Q.
if ((projected_rate <
target_rate * aq_c_transitions[aq_strength][i]) &&
if ((projected_rate < target_rate * aq_c_transitions[aq_strength][i]) &&
(logvar < (low_var_thresh + aq_c_var_thresholds[aq_strength][i]))) {
segment = i;
break;

5
vp9/encoder/vp9_aq_complexity.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_AQ_COMPLEXITY_H_
#define VP9_ENCODER_VP9_AQ_COMPLEXITY_H_
@@ -23,8 +22,8 @@ struct macroblock;
// Select a segment for the current Block.
void vp9_caq_select_segment(struct VP9_COMP *cpi, struct macroblock *,
BLOCK_SIZE bs,
int mi_row, int mi_col, int projected_rate);
BLOCK_SIZE bs, int mi_row, int mi_col,
int projected_rate);
// This function sets up a set of segments with delta Q values around
// the baseline frame quantizer.

164
vp9/encoder/vp9_aq_cyclicrefresh.c
View File

@@ -24,8 +24,7 @@
CYCLIC_REFRESH *vp9_cyclic_refresh_alloc(int mi_rows, int mi_cols) {
size_t last_coded_q_map_size;
CYCLIC_REFRESH *const cr = vpx_calloc(1, sizeof(*cr));
if (cr == NULL)
return NULL;
if (cr == NULL) return NULL;
cr->map = vpx_calloc(mi_rows * mi_cols, sizeof(*cr->map));
if (cr->map == NULL) {
@@ -53,11 +52,8 @@ void vp9_cyclic_refresh_free(CYCLIC_REFRESH *cr) {
// (lower-qp coding). Decision can be based on various factors, such as
// size of the coding block (i.e., below min_block size rejected), coding
// mode, and rate/distortion.
static int candidate_refresh_aq(const CYCLIC_REFRESH *cr,
const MODE_INFO *mi,
int64_t rate,
int64_t dist,
int bsize) {
static int candidate_refresh_aq(const CYCLIC_REFRESH *cr, const MODE_INFO *mi,
int64_t rate, int64_t dist, int bsize) {
MV mv = mi->mv[0].as_mv;
// Reject the block for lower-qp coding if projected distortion
// is above the threshold, and any of the following is true:
@@ -69,11 +65,9 @@ static int candidate_refresh_aq(const CYCLIC_REFRESH *cr,
mv.col > cr->motion_thresh || mv.col < -cr->motion_thresh ||
!is_inter_block(mi)))
return CR_SEGMENT_ID_BASE;
else if (bsize >= BLOCK_16X16 &&
rate < cr->thresh_rate_sb &&
is_inter_block(mi) &&
mi->mv[0].as_int == 0 &&
cr->rate_boost_fac > 10)
else if (bsize >= BLOCK_16X16 && rate < cr->thresh_rate_sb &&
is_inter_block(mi) && mi->mv[0].as_int == 0 &&
cr->rate_boost_fac > 10)
// More aggressive delta-q for bigger blocks with zero motion.
return CR_SEGMENT_ID_BOOST2;
else
@@ -84,9 +78,8 @@ static int candidate_refresh_aq(const CYCLIC_REFRESH *cr,
static int compute_deltaq(const VP9_COMP *cpi, int q, double rate_factor) {
const CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
const RATE_CONTROL *const rc = &cpi->rc;
int deltaq = vp9_compute_qdelta_by_rate(rc, cpi->common.frame_type,
q, rate_factor,
cpi->common.bit_depth);
int deltaq = vp9_compute_qdelta_by_rate(rc, cpi->common.frame_type, q,
rate_factor, cpi->common.bit_depth);
if ((-deltaq) > cr->max_qdelta_perc * q / 100) {
deltaq = -cr->max_qdelta_perc * q / 100;
}
@@ -109,17 +102,18 @@ int vp9_cyclic_refresh_estimate_bits_at_q(const VP9_COMP *cpi,
double weight_segment1 = (double)cr->actual_num_seg1_blocks / num8x8bl;
double weight_segment2 = (double)cr->actual_num_seg2_blocks / num8x8bl;
// Take segment weighted average for estimated bits.
estimated_bits = (int)((1.0 - weight_segment1 - weight_segment2) *
vp9_estimate_bits_at_q(cm->frame_type, cm->base_qindex, mbs,
correction_factor, cm->bit_depth) +
weight_segment1 *
vp9_estimate_bits_at_q(cm->frame_type,
cm->base_qindex + cr->qindex_delta[1], mbs,
correction_factor, cm->bit_depth) +
weight_segment2 *
vp9_estimate_bits_at_q(cm->frame_type,
cm->base_qindex + cr->qindex_delta[2], mbs,
correction_factor, cm->bit_depth));
estimated_bits =
(int)((1.0 - weight_segment1 - weight_segment2) *
vp9_estimate_bits_at_q(cm->frame_type, cm->base_qindex, mbs,
correction_factor, cm->bit_depth) +
weight_segment1 *
vp9_estimate_bits_at_q(cm->frame_type,
cm->base_qindex + cr->qindex_delta[1],
mbs, correction_factor, cm->bit_depth) +
weight_segment2 *
vp9_estimate_bits_at_q(cm->frame_type,
cm->base_qindex + cr->qindex_delta[2],
mbs, correction_factor, cm->bit_depth));
return estimated_bits;
}
@@ -137,30 +131,29 @@ int vp9_cyclic_refresh_rc_bits_per_mb(const VP9_COMP *cpi, int i,
// Weight for segment prior to encoding: take the average of the target
// number for the frame to be encoded and the actual from the previous frame.
int target_refresh = cr->percent_refresh * cm->mi_rows * cm->mi_cols / 100;
double weight_segment = (double)((target_refresh +
cr->actual_num_seg1_blocks + cr->actual_num_seg2_blocks) >> 1) /
double weight_segment =
(double)((target_refresh + cr->actual_num_seg1_blocks +
cr->actual_num_seg2_blocks) >>
1) /
num8x8bl;
// Compute delta-q corresponding to qindex i.
int deltaq = compute_deltaq(cpi, i, cr->rate_ratio_qdelta);
// Take segment weighted average for bits per mb.
bits_per_mb = (int)((1.0 - weight_segment) *
vp9_rc_bits_per_mb(cm->frame_type, i, correction_factor, cm->bit_depth) +
weight_segment *
vp9_rc_bits_per_mb(cm->frame_type, i + deltaq, correction_factor,
cm->bit_depth));
vp9_rc_bits_per_mb(cm->frame_type, i,
correction_factor, cm->bit_depth) +
weight_segment *
vp9_rc_bits_per_mb(cm->frame_type, i + deltaq,
correction_factor, cm->bit_depth));
return bits_per_mb;
}
// Prior to coding a given prediction block, of size bsize at (mi_row, mi_col),
// check if we should reset the segment_id, and update the cyclic_refresh map
// and segmentation map.
void vp9_cyclic_refresh_update_segment(VP9_COMP *const cpi,
MODE_INFO *const mi,
int mi_row, int mi_col,
BLOCK_SIZE bsize,
int64_t rate,
int64_t dist,
int skip,
void vp9_cyclic_refresh_update_segment(VP9_COMP *const cpi, MODE_INFO *const mi,
int mi_row, int mi_col, BLOCK_SIZE bsize,
int64_t rate, int64_t dist, int skip,
struct macroblock_plane *const p) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
@@ -172,22 +165,16 @@ void vp9_cyclic_refresh_update_segment(VP9_COMP *const cpi,
int refresh_this_block = candidate_refresh_aq(cr, mi, rate, dist, bsize);
// Default is to not update the refresh map.
int new_map_value = cr->map[block_index];
int x = 0; int y = 0;
int x = 0;
int y = 0;
int is_skin = 0;
if (refresh_this_block == 0 &&
bsize <= BLOCK_16X16 &&
if (refresh_this_block == 0 && bsize <= BLOCK_16X16 &&
cpi->use_skin_detection) {
is_skin = vp9_compute_skin_block(p[0].src.buf,
p[1].src.buf,
p[2].src.buf,
p[0].src.stride,
p[1].src.stride,
bsize,
0,
0);
if (is_skin)
refresh_this_block = 1;
is_skin =
vp9_compute_skin_block(p[0].src.buf, p[1].src.buf, p[2].src.buf,
p[0].src.stride, p[1].src.stride, bsize, 0, 0);
if (is_skin) refresh_this_block = 1;
}
if (cpi->oxcf.rc_mode == VPX_VBR && mi->ref_frame[0] == GOLDEN_FRAME)
@@ -198,8 +185,7 @@ void vp9_cyclic_refresh_update_segment(VP9_COMP *const cpi,
if (cyclic_refresh_segment_id_boosted(mi->segment_id)) {
mi->segment_id = refresh_this_block;
// Reset segment_id if it will be skipped.
if (skip)
mi->segment_id = CR_SEGMENT_ID_BASE;
if (skip) mi->segment_id = CR_SEGMENT_ID_BASE;
}
// Update the cyclic refresh map, to be used for setting segmentation map
@@ -212,8 +198,7 @@ void vp9_cyclic_refresh_update_segment(VP9_COMP *const cpi,
// Else if it is accepted as candidate for refresh, and has not already
// been refreshed (marked as 1) then mark it as a candidate for cleanup
// for future time (marked as 0), otherwise don't update it.
if (cr->map[block_index] == 1)
new_map_value = 0;
if (cr->map[block_index] == 1) new_map_value = 0;
} else {
// Leave it marked as block that is not candidate for refresh.
new_map_value = 1;
@@ -250,13 +235,12 @@ void vp9_cyclic_refresh_update_sb_postencode(VP9_COMP *const cpi,
// the map for this spatial location is not entirely correct.
if ((!is_inter_block(mi) || !mi->skip) &&
mi->segment_id <= CR_SEGMENT_ID_BOOST2) {
cr->last_coded_q_map[map_offset] = clamp(
cm->base_qindex + cr->qindex_delta[mi->segment_id], 0, MAXQ);
cr->last_coded_q_map[map_offset] =
clamp(cm->base_qindex + cr->qindex_delta[mi->segment_id], 0, MAXQ);
} else if (is_inter_block(mi) && mi->skip &&
mi->segment_id <= CR_SEGMENT_ID_BOOST2) {
cr->last_coded_q_map[map_offset] = VPXMIN(
clamp(cm->base_qindex + cr->qindex_delta[mi->segment_id],
0, MAXQ),
clamp(cm->base_qindex + cr->qindex_delta[mi->segment_id], 0, MAXQ),
cr->last_coded_q_map[map_offset]);
}
}
@@ -272,11 +256,12 @@ void vp9_cyclic_refresh_postencode(VP9_COMP *const cpi) {
cr->actual_num_seg2_blocks = 0;
for (mi_row = 0; mi_row < cm->mi_rows; mi_row++)
for (mi_col = 0; mi_col < cm->mi_cols; mi_col++) {
if (cyclic_refresh_segment_id(
seg_map[mi_row * cm->mi_cols + mi_col]) == CR_SEGMENT_ID_BOOST1)
if (cyclic_refresh_segment_id(seg_map[mi_row * cm->mi_cols + mi_col]) ==
CR_SEGMENT_ID_BOOST1)
cr->actual_num_seg1_blocks++;
else if (cyclic_refresh_segment_id(
seg_map[mi_row * cm->mi_cols + mi_col]) == CR_SEGMENT_ID_BOOST2)
seg_map[mi_row * cm->mi_cols + mi_col]) ==
CR_SEGMENT_ID_BOOST2)
cr->actual_num_seg2_blocks++;
}
}
@@ -292,8 +277,7 @@ void vp9_cyclic_refresh_set_golden_update(VP9_COMP *const cpi) {
rc->baseline_gf_interval = VPXMIN(4 * (100 / cr->percent_refresh), 40);
else
rc->baseline_gf_interval = 40;
if (cpi->oxcf.rc_mode == VPX_VBR)
rc->baseline_gf_interval = 20;
if (cpi->oxcf.rc_mode == VPX_VBR) rc->baseline_gf_interval = 20;
}
// Update some encoding stats (from the just encoded frame). If this frame's
@@ -315,21 +299,21 @@ void vp9_cyclic_refresh_check_golden_update(VP9_COMP *const cpi) {
for (mi_row = 0; mi_row < rows; mi_row++) {
for (mi_col = 0; mi_col < cols; mi_col++) {
if (flag_force_gf_high_motion == 1) {
int16_t abs_mvr = mi[0]->mv[0].as_mv.row >= 0 ?
mi[0]->mv[0].as_mv.row : -1 * mi[0]->mv[0].as_mv.row;
int16_t abs_mvc = mi[0]->mv[0].as_mv.col >= 0 ?
mi[0]->mv[0].as_mv.col : -1 * mi[0]->mv[0].as_mv.col;
int16_t abs_mvr = mi[0]->mv[0].as_mv.row >= 0
? mi[0]->mv[0].as_mv.row
: -1 * mi[0]->mv[0].as_mv.row;
int16_t abs_mvc = mi[0]->mv[0].as_mv.col >= 0
? mi[0]->mv[0].as_mv.col
: -1 * mi[0]->mv[0].as_mv.col;
// Calculate the motion of the background.
if (abs_mvr <= 16 && abs_mvc <= 16) {
cnt1++;
if (abs_mvr == 0 && abs_mvc == 0)
cnt2++;
if (abs_mvr == 0 && abs_mvc == 0) cnt2++;
}
}
mi++;
// Accumulate low_content_frame.
if (cr->map[mi_row * cols + mi_col] < 1)
low_content_frame++;
if (cr->map[mi_row * cols + mi_col] < 1) low_content_frame++;
}
mi += 8;
}
@@ -339,7 +323,7 @@ void vp9_cyclic_refresh_check_golden_update(VP9_COMP *const cpi) {
// Also, force this frame as a golden update frame if this frame will change
// the resolution (resize_pending != 0).
if (cpi->resize_pending != 0 ||
(cnt1 * 100 > (70 * rows * cols) && cnt2 * 20 < cnt1)) {
(cnt1 * 100 > (70 * rows * cols) && cnt2 * 20 < cnt1)) {
vp9_cyclic_refresh_set_golden_update(cpi);
rc->frames_till_gf_update_due = rc->baseline_gf_interval;
@@ -348,8 +332,7 @@ void vp9_cyclic_refresh_check_golden_update(VP9_COMP *const cpi) {
cpi->refresh_golden_frame = 1;
force_gf_refresh = 1;
}
fraction_low =
(double)low_content_frame / (rows * cols);
fraction_low = (double)low_content_frame / (rows * cols);
// Update average.
cr->low_content_avg = (fraction_low + 3 * cr->low_content_avg) / 4;
if (!force_gf_refresh && cpi->refresh_golden_frame == 1) {
@@ -396,14 +379,14 @@ static void cyclic_refresh_update_map(VP9_COMP *const cpi) {
}
qindex_thresh =
cpi->oxcf.content == VP9E_CONTENT_SCREEN
? vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST2, cm->base_qindex)
: vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST1, cm->base_qindex);
? vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST2, cm->base_qindex)
: vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST1, cm->base_qindex);
// More aggressive settings for noisy content.
if (cpi->noise_estimate.enabled && cpi->noise_estimate.level >= kMedium) {
consec_zero_mv_thresh = 80;
qindex_thresh =
VPXMAX(vp9_get_qindex(&cm->seg, CR_SEGMENT_ID_BOOST1, cm->base_qindex),
7 * cm->base_qindex >> 3);
7 * cm->base_qindex >> 3);
}
do {
int sum_map = 0;
@@ -454,8 +437,7 @@ static void cyclic_refresh_update_map(VP9_COMP *const cpi) {
} while (cr->target_num_seg_blocks < block_count && i != cr->sb_index);
cr->sb_index = i;
cr->reduce_refresh = 0;
if (count_sel < (3 * count_tot) >> 2)
cr->reduce_refresh = 1;
if (count_sel<(3 * count_tot)>> 2) cr->reduce_refresh = 1;
}
// Set cyclic refresh parameters.
@@ -464,8 +446,7 @@ void vp9_cyclic_refresh_update_parameters(VP9_COMP *const cpi) {
const VP9_COMMON *const cm = &cpi->common;
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
cr->percent_refresh = 10;
if (cr->reduce_refresh)
cr->percent_refresh = 5;
if (cr->reduce_refresh) cr->percent_refresh = 5;
cr->max_qdelta_perc = 50;
cr->time_for_refresh = 0;
cr->motion_thresh = 32;
@@ -474,8 +455,8 @@ void vp9_cyclic_refresh_update_parameters(VP9_COMP *const cpi) {
// periods of the refresh cycle, after a key frame.
// Account for larger interval on base layer for temporal layers.
if (cr->percent_refresh > 0 &&
rc->frames_since_key < (4 * cpi->svc.number_temporal_layers) *
(100 / cr->percent_refresh)) {
rc->frames_since_key <
(4 * cpi->svc.number_temporal_layers) * (100 / cr->percent_refresh)) {
cr->rate_ratio_qdelta = 3.0;
} else {
cr->rate_ratio_qdelta = 2.0;
@@ -486,9 +467,7 @@ void vp9_cyclic_refresh_update_parameters(VP9_COMP *const cpi) {
}
}
// Adjust some parameters for low resolutions at low bitrates.
if (cm->width <= 352 &&
cm->height <= 288 &&
rc->avg_frame_bandwidth < 3400) {
if (cm->width <= 352 && cm->height <= 288 && rc->avg_frame_bandwidth < 3400) {
cr->motion_thresh = 4;
cr->rate_boost_fac = 10;
}
@@ -520,11 +499,9 @@ void vp9_cyclic_refresh_setup(VP9_COMP *const cpi) {
// instead of completely shutting off at low bitrates. For now keep it on.
// const int apply_cyclic_refresh = apply_cyclic_refresh_bitrate(cm, rc);
const int apply_cyclic_refresh = 1;
if (cm->current_video_frame == 0)
cr->low_content_avg = 0.0;
if (cm->current_video_frame == 0) cr->low_content_avg = 0.0;
// Don't apply refresh on key frame or temporal enhancement layer frames.
if (!apply_cyclic_refresh ||
(cm->frame_type == KEY_FRAME) ||
if (!apply_cyclic_refresh || (cm->frame_type == KEY_FRAME) ||
(cpi->svc.temporal_layer_id > 0)) {
// Set segmentation map to 0 and disable.
unsigned char *const seg_map = cpi->segmentation_map;
@@ -590,8 +567,7 @@ void vp9_cyclic_refresh_setup(VP9_COMP *const cpi) {
vp9_set_segdata(seg, CR_SEGMENT_ID_BOOST2, SEG_LVL_ALT_Q, qindex_delta);
// Reset if resoluton change has occurred.
if (cpi->resize_pending != 0)
vp9_cyclic_refresh_reset_resize(cpi);
if (cpi->resize_pending != 0) vp9_cyclic_refresh_reset_resize(cpi);
// Update the segmentation and refresh map.
cyclic_refresh_update_map(cpi);

11
vp9/encoder/vp9_aq_cyclicrefresh.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_AQ_CYCLICREFRESH_H_
#define VP9_ENCODER_VP9_AQ_CYCLICREFRESH_H_
@@ -23,9 +22,9 @@ extern "C" {
// The segment ids used in cyclic refresh: from base (no boost) to increasing
// boost (higher delta-qp).
#define CR_SEGMENT_ID_BASE 0
#define CR_SEGMENT_ID_BOOST1 1
#define CR_SEGMENT_ID_BOOST2 2
#define CR_SEGMENT_ID_BASE 0
#define CR_SEGMENT_ID_BOOST1 1
#define CR_SEGMENT_ID_BOOST2 2
// Maximum rate target ratio for setting segment delta-qp.
#define CR_MAX_RATE_TARGET_RATIO 4.0
@@ -91,8 +90,8 @@ int vp9_cyclic_refresh_rc_bits_per_mb(const struct VP9_COMP *cpi, int i,
// check if we should reset the segment_id, and update the cyclic_refresh map
// and segmentation map.
void vp9_cyclic_refresh_update_segment(struct VP9_COMP *const cpi,
MODE_INFO *const mi,
int mi_row, int mi_col, BLOCK_SIZE bsize,
MODE_INFO *const mi, int mi_row,
int mi_col, BLOCK_SIZE bsize,
int64_t rate, int64_t dist, int skip,
struct macroblock_plane *const p);

65
vp9/encoder/vp9_aq_variance.c
View File

@@ -23,19 +23,19 @@
#define ENERGY_MIN (-4)
#define ENERGY_MAX (1)
#define ENERGY_SPAN (ENERGY_MAX - ENERGY_MIN + 1)
#define ENERGY_IN_BOUNDS(energy)\
#define ENERGY_SPAN (ENERGY_MAX - ENERGY_MIN + 1)
#define ENERGY_IN_BOUNDS(energy) \
assert((energy) >= ENERGY_MIN && (energy) <= ENERGY_MAX)
static const double rate_ratio[MAX_SEGMENTS] =
{2.5, 2.0, 1.5, 1.0, 0.75, 1.0, 1.0, 1.0};
static const int segment_id[ENERGY_SPAN] = {0, 1, 1, 2, 3, 4};
static const double rate_ratio[MAX_SEGMENTS] = { 2.5, 2.0, 1.5, 1.0,
0.75, 1.0, 1.0, 1.0 };
static const int segment_id[ENERGY_SPAN] = { 0, 1, 1, 2, 3, 4 };
#define SEGMENT_ID(i) segment_id[(i) - ENERGY_MIN]
#define SEGMENT_ID(i) segment_id[(i)-ENERGY_MIN]
DECLARE_ALIGNED(16, static const uint8_t, vp9_64_zeros[64]) = {0};
DECLARE_ALIGNED(16, static const uint8_t, vp9_64_zeros[64]) = { 0 };
#if CONFIG_VP9_HIGHBITDEPTH
DECLARE_ALIGNED(16, static const uint16_t, vp9_highbd_64_zeros[64]) = {0};
DECLARE_ALIGNED(16, static const uint16_t, vp9_highbd_64_zeros[64]) = { 0 };
#endif
unsigned int vp9_vaq_segment_id(int energy) {
@@ -85,9 +85,9 @@ void vp9_vaq_frame_setup(VP9_COMP *cpi) {
/* TODO(agrange, paulwilkins): The block_variance calls the unoptimized versions
* of variance() and highbd_8_variance(). It should not.
*/
static void aq_variance(const uint8_t *a, int a_stride,
const uint8_t *b, int b_stride,
int w, int h, unsigned int *sse, int *sum) {
static void aq_variance(const uint8_t *a, int a_stride, const uint8_t *b,
int b_stride, int w, int h, unsigned int *sse,
int *sum) {
int i, j;
*sum = 0;
@@ -106,9 +106,9 @@ static void aq_variance(const uint8_t *a, int a_stride,
}
#if CONFIG_VP9_HIGHBITDEPTH
static void aq_highbd_variance64(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, uint64_t *sse, uint64_t *sum) {
static void aq_highbd_variance64(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride, int w, int h,
uint64_t *sse, uint64_t *sum) {
int i, j;
uint16_t *a = CONVERT_TO_SHORTPTR(a8);
@@ -127,9 +127,9 @@ static void aq_highbd_variance64(const uint8_t *a8, int a_stride,
}
}
static void aq_highbd_8_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, unsigned int *sse, int *sum) {
static void aq_highbd_8_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride, int w, int h,
unsigned int *sse, int *sum) {
uint64_t sse_long = 0;
uint64_t sum_long = 0;
aq_highbd_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
@@ -142,10 +142,10 @@ static unsigned int block_variance(VP9_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bs) {
MACROBLOCKD *xd = &x->e_mbd;
unsigned int var, sse;
int right_overflow = (xd->mb_to_right_edge < 0) ?
((-xd->mb_to_right_edge) >> 3) : 0;
int bottom_overflow = (xd->mb_to_bottom_edge < 0) ?
((-xd->mb_to_bottom_edge) >> 3) : 0;
int right_overflow =
(xd->mb_to_right_edge < 0) ? ((-xd->mb_to_right_edge) >> 3) : 0;
int bottom_overflow =
(xd->mb_to_bottom_edge < 0) ? ((-xd->mb_to_bottom_edge) >> 3) : 0;
if (right_overflow || bottom_overflow) {
const int bw = 8 * num_8x8_blocks_wide_lookup[bs] - right_overflow;
@@ -159,30 +159,27 @@ static unsigned int block_variance(VP9_COMP *cpi, MACROBLOCK *x,
sse >>= 2 * (xd->bd - 8);
avg >>= (xd->bd - 8);
} else {
aq_variance(x->plane[0].src.buf, x->plane[0].src.stride,
vp9_64_zeros, 0, bw, bh, &sse, &avg);
aq_variance(x->plane[0].src.buf, x->plane[0].src.stride, vp9_64_zeros, 0,
bw, bh, &sse, &avg);
}
#else
aq_variance(x->plane[0].src.buf, x->plane[0].src.stride,
vp9_64_zeros, 0, bw, bh, &sse, &avg);
aq_variance(x->plane[0].src.buf, x->plane[0].src.stride, vp9_64_zeros, 0,
bw, bh, &sse, &avg);
#endif // CONFIG_VP9_HIGHBITDEPTH
var = sse - (((int64_t)avg * avg) / (bw * bh));
return (unsigned int)(((uint64_t)256 * var) / (bw * bh));
} else {
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
var = cpi->fn_ptr[bs].vf(x->plane[0].src.buf,
x->plane[0].src.stride,
CONVERT_TO_BYTEPTR(vp9_highbd_64_zeros),
0, &sse);
var =
cpi->fn_ptr[bs].vf(x->plane[0].src.buf, x->plane[0].src.stride,
CONVERT_TO_BYTEPTR(vp9_highbd_64_zeros), 0, &sse);
} else {
var = cpi->fn_ptr[bs].vf(x->plane[0].src.buf,
x->plane[0].src.stride,
var = cpi->fn_ptr[bs].vf(x->plane[0].src.buf, x->plane[0].src.stride,
vp9_64_zeros, 0, &sse);
}
#else
var = cpi->fn_ptr[bs].vf(x->plane[0].src.buf,
x->plane[0].src.stride,
var = cpi->fn_ptr[bs].vf(x->plane[0].src.buf, x->plane[0].src.stride,
vp9_64_zeros, 0, &sse);
#endif // CONFIG_VP9_HIGHBITDEPTH
return (unsigned int)(((uint64_t)256 * var) >> num_pels_log2_lookup[bs]);
@@ -201,7 +198,7 @@ int vp9_block_energy(VP9_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bs) {
double energy_midpoint;
vpx_clear_system_state();
energy_midpoint =
(cpi->oxcf.pass == 2) ? cpi->twopass.mb_av_energy : DEFAULT_E_MIDPOINT;
(cpi->oxcf.pass == 2) ? cpi->twopass.mb_av_energy : DEFAULT_E_MIDPOINT;
energy = vp9_log_block_var(cpi, x, bs) - energy_midpoint;
return clamp((int)round(energy), ENERGY_MIN, ENERGY_MAX);
}

1
vp9/encoder/vp9_aq_variance.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_AQ_VARIANCE_H_
#define VP9_ENCODER_VP9_AQ_VARIANCE_H_

221
vp9/encoder/vp9_bitstream.c
View File

@@ -36,14 +36,17 @@
#include "vp9/encoder/vp9_tokenize.h"
static const struct vp9_token intra_mode_encodings[INTRA_MODES] = {
{0, 1}, {6, 3}, {28, 5}, {30, 5}, {58, 6}, {59, 6}, {126, 7}, {127, 7},
{62, 6}, {2, 2}};
{ 0, 1 }, { 6, 3 }, { 28, 5 }, { 30, 5 }, { 58, 6 },
{ 59, 6 }, { 126, 7 }, { 127, 7 }, { 62, 6 }, { 2, 2 }
};
static const struct vp9_token switchable_interp_encodings[SWITCHABLE_FILTERS] =
{{0, 1}, {2, 2}, {3, 2}};
static const struct vp9_token partition_encodings[PARTITION_TYPES] =
{{0, 1}, {2, 2}, {6, 3}, {7, 3}};
static const struct vp9_token inter_mode_encodings[INTER_MODES] =
{{2, 2}, {6, 3}, {0, 1}, {7, 3}};
{ { 0, 1 }, { 2, 2 }, { 3, 2 } };
static const struct vp9_token partition_encodings[PARTITION_TYPES] = {
{ 0, 1 }, { 2, 2 }, { 6, 3 }, { 7, 3 }
};
static const struct vp9_token inter_mode_encodings[INTER_MODES] = {
{ 2, 2 }, { 6, 3 }, { 0, 1 }, { 7, 3 }
};
static void write_intra_mode(vpx_writer *w, PREDICTION_MODE mode,
const vpx_prob *probs) {
@@ -57,15 +60,15 @@ static void write_inter_mode(vpx_writer *w, PREDICTION_MODE mode,
&inter_mode_encodings[INTER_OFFSET(mode)]);
}
static void encode_unsigned_max(struct vpx_write_bit_buffer *wb,
int data, int max) {
static void encode_unsigned_max(struct vpx_write_bit_buffer *wb, int data,
int max) {
vpx_wb_write_literal(wb, data, get_unsigned_bits(max));
}
static void prob_diff_update(const vpx_tree_index *tree,
vpx_prob probs[/*n - 1*/],
const unsigned int counts[/*n - 1*/],
int n, vpx_writer *w) {
const unsigned int counts[/*n - 1*/], int n,
vpx_writer *w) {
int i;
unsigned int branch_ct[32][2];
@@ -77,13 +80,13 @@ static void prob_diff_update(const vpx_tree_index *tree,
vp9_cond_prob_diff_update(w, &probs[i], branch_ct[i]);
}
static void write_selected_tx_size(const VP9_COMMON *cm,
const MACROBLOCKD *xd, vpx_writer *w) {
static void write_selected_tx_size(const VP9_COMMON *cm, const MACROBLOCKD *xd,
vpx_writer *w) {
TX_SIZE tx_size = xd->mi[0]->tx_size;
BLOCK_SIZE bsize = xd->mi[0]->sb_type;
const TX_SIZE max_tx_size = max_txsize_lookup[bsize];
const vpx_prob *const tx_probs = get_tx_probs2(max_tx_size, xd,
&cm->fc->tx_probs);
const vpx_prob *const tx_probs =
get_tx_probs2(max_tx_size, xd, &cm->fc->tx_probs);
vpx_write(w, tx_size != TX_4X4, tx_probs[0]);
if (tx_size != TX_4X4 && max_tx_size >= TX_16X16) {
vpx_write(w, tx_size != TX_8X8, tx_probs[1]);
@@ -120,18 +123,18 @@ static void update_switchable_interp_probs(VP9_COMMON *cm, vpx_writer *w,
counts->switchable_interp[j], SWITCHABLE_FILTERS, w);
}
static void pack_mb_tokens(vpx_writer *w,
TOKENEXTRA **tp, const TOKENEXTRA *const stop,
static void pack_mb_tokens(vpx_writer *w, TOKENEXTRA **tp,
const TOKENEXTRA *const stop,
vpx_bit_depth_t bit_depth) {
const TOKENEXTRA *p;
const vp9_extra_bit *const extra_bits =
#if CONFIG_VP9_HIGHBITDEPTH
(bit_depth == VPX_BITS_12) ? vp9_extra_bits_high12 :
(bit_depth == VPX_BITS_10) ? vp9_extra_bits_high10 :
vp9_extra_bits;
(bit_depth == VPX_BITS_12)
? vp9_extra_bits_high12
: (bit_depth == VPX_BITS_10) ? vp9_extra_bits_high10 : vp9_extra_bits;
#else
vp9_extra_bits;
(void) bit_depth;
vp9_extra_bits;
(void)bit_depth;
#endif // CONFIG_VP9_HIGHBITDEPTH
for (p = *tp; p < stop && p->token != EOSB_TOKEN; ++p) {
@@ -144,7 +147,7 @@ static void pack_mb_tokens(vpx_writer *w,
vpx_write(w, 0, p->context_tree[1]);
++p;
if (p == stop || p->token == EOSB_TOKEN) {
*tp = (TOKENEXTRA*)(uintptr_t)p + (p->token == EOSB_TOKEN);
*tp = (TOKENEXTRA *)(uintptr_t)p + (p->token == EOSB_TOKEN);
return;
}
}
@@ -182,7 +185,7 @@ static void pack_mb_tokens(vpx_writer *w,
}
}
}
*tp = (TOKENEXTRA*)(uintptr_t)p + (p->token == EOSB_TOKEN);
*tp = (TOKENEXTRA *)(uintptr_t)p + (p->token == EOSB_TOKEN);
}
static void write_segment_id(vpx_writer *w, const struct segmentation *seg,
@@ -203,7 +206,7 @@ static void write_ref_frames(const VP9_COMMON *cm, const MACROBLOCKD *xd,
if (segfeature_active(&cm->seg, segment_id, SEG_LVL_REF_FRAME)) {
assert(!is_compound);
assert(mi->ref_frame[0] ==
get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME));
get_segdata(&cm->seg, segment_id, SEG_LVL_REF_FRAME));
} else {
// does the feature use compound prediction or not
// (if not specified at the frame/segment level)
@@ -248,8 +251,7 @@ static void pack_inter_mode_mvs(VP9_COMP *cpi, const MODE_INFO *mi,
const int pred_flag = mi->seg_id_predicted;
vpx_prob pred_prob = vp9_get_pred_prob_seg_id(seg, xd);
vpx_write(w, pred_flag, pred_prob);
if (!pred_flag)
write_segment_id(w, seg, segment_id);
if (!pred_flag) write_segment_id(w, seg, segment_id);
} else {
write_segment_id(w, seg, segment_id);
}
@@ -338,8 +340,7 @@ static void write_mb_modes_kf(const VP9_COMMON *cm, const MACROBLOCKD *xd,
const MODE_INFO *const left_mi = xd->left_mi;
const BLOCK_SIZE bsize = mi->sb_type;
if (seg->update_map)
write_segment_id(w, seg, mi->segment_id);
if (seg->update_map) write_segment_id(w, seg, mi->segment_id);
write_skip(cm, xd, mi->segment_id, mi, w);
@@ -367,8 +368,8 @@ static void write_mb_modes_kf(const VP9_COMMON *cm, const MACROBLOCKD *xd,
static void write_modes_b(VP9_COMP *cpi, const TileInfo *const tile,
vpx_writer *w, TOKENEXTRA **tok,
const TOKENEXTRA *const tok_end,
int mi_row, int mi_col) {
const TOKENEXTRA *const tok_end, int mi_row,
int mi_col) {
const VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
MODE_INFO *m;
@@ -376,13 +377,12 @@ static void write_modes_b(VP9_COMP *cpi, const TileInfo *const tile,
xd->mi = cm->mi_grid_visible + (mi_row * cm->mi_stride + mi_col);
m = xd->mi[0];
cpi->td.mb.mbmi_ext = cpi->td.mb.mbmi_ext_base +
(mi_row * cm->mi_cols + mi_col);
cpi->td.mb.mbmi_ext =
cpi->td.mb.mbmi_ext_base + (mi_row * cm->mi_cols + mi_col);
set_mi_row_col(xd, tile,
mi_row, num_8x8_blocks_high_lookup[m->sb_type],
mi_col, num_8x8_blocks_wide_lookup[m->sb_type],
cm->mi_rows, cm->mi_cols);
set_mi_row_col(xd, tile, mi_row, num_8x8_blocks_high_lookup[m->sb_type],
mi_col, num_8x8_blocks_wide_lookup[m->sb_type], cm->mi_rows,
cm->mi_cols);
if (frame_is_intra_only(cm)) {
write_mb_modes_kf(cm, xd, xd->mi, w);
} else {
@@ -394,9 +394,9 @@ static void write_modes_b(VP9_COMP *cpi, const TileInfo *const tile,
}
static void write_partition(const VP9_COMMON *const cm,
const MACROBLOCKD *const xd,
int hbs, int mi_row, int mi_col,
PARTITION_TYPE p, BLOCK_SIZE bsize, vpx_writer *w) {
const MACROBLOCKD *const xd, int hbs, int mi_row,
int mi_col, PARTITION_TYPE p, BLOCK_SIZE bsize,
vpx_writer *w) {
const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
const vpx_prob *const probs = xd->partition_probs[ctx];
const int has_rows = (mi_row + hbs) < cm->mi_rows;
@@ -415,10 +415,10 @@ static void write_partition(const VP9_COMMON *const cm,
}
}
static void write_modes_sb(VP9_COMP *cpi,
const TileInfo *const tile, vpx_writer *w,
TOKENEXTRA **tok, const TOKENEXTRA *const tok_end,
int mi_row, int mi_col, BLOCK_SIZE bsize) {
static void write_modes_sb(VP9_COMP *cpi, const TileInfo *const tile,
vpx_writer *w, TOKENEXTRA **tok,
const TOKENEXTRA *const tok_end, int mi_row,
int mi_col, BLOCK_SIZE bsize) {
const VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
@@ -428,8 +428,7 @@ static void write_modes_sb(VP9_COMP *cpi,
BLOCK_SIZE subsize;
const MODE_INFO *m = NULL;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
return;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
m = cm->mi_grid_visible[mi_row * cm->mi_stride + mi_col];
@@ -462,8 +461,7 @@ static void write_modes_sb(VP9_COMP *cpi,
write_modes_sb(cpi, tile, w, tok, tok_end, mi_row + bs, mi_col + bs,
subsize);
break;
default:
assert(0);
default: assert(0);
}
}
@@ -473,9 +471,9 @@ static void write_modes_sb(VP9_COMP *cpi,
update_partition_context(xd, mi_row, mi_col, subsize, bsize);
}
static void write_modes(VP9_COMP *cpi,
const TileInfo *const tile, vpx_writer *w,
TOKENEXTRA **tok, const TOKENEXTRA *const tok_end) {
static void write_modes(VP9_COMP *cpi, const TileInfo *const tile,
vpx_writer *w, TOKENEXTRA **tok,
const TOKENEXTRA *const tok_end) {
const VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
int mi_row, mi_col;
@@ -487,8 +485,7 @@ static void write_modes(VP9_COMP *cpi,
vp9_zero(xd->left_seg_context);
for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE)
write_modes_sb(cpi, tile, w, tok, tok_end, mi_row, mi_col,
BLOCK_64X64);
write_modes_sb(cpi, tile, w, tok, tok_end, mi_row, mi_col, BLOCK_64X64);
}
}
@@ -496,7 +493,7 @@ static void build_tree_distribution(VP9_COMP *cpi, TX_SIZE tx_size,
vp9_coeff_stats *coef_branch_ct,
vp9_coeff_probs_model *coef_probs) {
vp9_coeff_count *coef_counts = cpi->td.rd_counts.coef_counts[tx_size];
unsigned int (*eob_branch_ct)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] =
unsigned int(*eob_branch_ct)[REF_TYPES][COEF_BANDS][COEFF_CONTEXTS] =
cpi->common.counts.eob_branch[tx_size];
int i, j, k, l, m;
@@ -507,19 +504,19 @@ static void build_tree_distribution(VP9_COMP *cpi, TX_SIZE tx_size,
vp9_tree_probs_from_distribution(vp9_coef_tree,
coef_branch_ct[i][j][k][l],
coef_counts[i][j][k][l]);
coef_branch_ct[i][j][k][l][0][1] = eob_branch_ct[i][j][k][l] -
coef_branch_ct[i][j][k][l][0][0];
coef_branch_ct[i][j][k][l][0][1] =
eob_branch_ct[i][j][k][l] - coef_branch_ct[i][j][k][l][0][0];
for (m = 0; m < UNCONSTRAINED_NODES; ++m)
coef_probs[i][j][k][l][m] = get_binary_prob(
coef_branch_ct[i][j][k][l][m][0],
coef_branch_ct[i][j][k][l][m][1]);
coef_probs[i][j][k][l][m] =
get_binary_prob(coef_branch_ct[i][j][k][l][m][0],
coef_branch_ct[i][j][k][l][m][1]);
}
}
}
}
}
static void update_coef_probs_common(vpx_writer* const bc, VP9_COMP *cpi,
static void update_coef_probs_common(vpx_writer *const bc, VP9_COMP *cpi,
TX_SIZE tx_size,
vp9_coeff_stats *frame_branch_ct,
vp9_coeff_probs_model *new_coef_probs) {
@@ -533,7 +530,7 @@ static void update_coef_probs_common(vpx_writer* const bc, VP9_COMP *cpi,
case TWO_LOOP: {
/* dry run to see if there is any update at all needed */
int savings = 0;
int update[2] = {0, 0};
int update[2] = { 0, 0 };
for (i = 0; i < PLANE_TYPES; ++i) {
for (j = 0; j < REF_TYPES; ++j) {
for (k = 0; k < COEF_BANDS; ++k) {
@@ -550,8 +547,7 @@ static void update_coef_probs_common(vpx_writer* const bc, VP9_COMP *cpi,
else
s = vp9_prob_diff_update_savings_search(
frame_branch_ct[i][j][k][l][t], oldp, &newp, upd);
if (s > 0 && newp != oldp)
u = 1;
if (s > 0 && newp != oldp) u = 1;
if (u)
savings += s - (int)(vp9_cost_zero(upd));
else
@@ -583,14 +579,12 @@ static void update_coef_probs_common(vpx_writer* const bc, VP9_COMP *cpi,
int u = 0;
if (t == PIVOT_NODE)
s = vp9_prob_diff_update_savings_search_model(
frame_branch_ct[i][j][k][l][0],
*oldp, &newp, upd, stepsize);
frame_branch_ct[i][j][k][l][0], *oldp, &newp, upd,
stepsize);
else
s = vp9_prob_diff_update_savings_search(
frame_branch_ct[i][j][k][l][t],
*oldp, &newp, upd);
if (s > 0 && newp != *oldp)
u = 1;
frame_branch_ct[i][j][k][l][t], *oldp, &newp, upd);
if (s > 0 && newp != *oldp) u = 1;
vpx_write(bc, u, upd);
if (u) {
/* send/use new probability */
@@ -621,16 +615,14 @@ static void update_coef_probs_common(vpx_writer* const bc, VP9_COMP *cpi,
if (t == PIVOT_NODE) {
s = vp9_prob_diff_update_savings_search_model(
frame_branch_ct[i][j][k][l][0],
*oldp, &newp, upd, stepsize);
frame_branch_ct[i][j][k][l][0], *oldp, &newp, upd,
stepsize);
} else {
s = vp9_prob_diff_update_savings_search(
frame_branch_ct[i][j][k][l][t],
*oldp, &newp, upd);
frame_branch_ct[i][j][k][l][t], *oldp, &newp, upd);
}
if (s > 0 && newp != *oldp)
u = 1;
if (s > 0 && newp != *oldp) u = 1;
updates += u;
if (u == 0 && updates == 0) {
noupdates_before_first++;
@@ -659,12 +651,11 @@ static void update_coef_probs_common(vpx_writer* const bc, VP9_COMP *cpi,
}
return;
}
default:
assert(0);
default: assert(0);
}
}
static void update_coef_probs(VP9_COMP *cpi, vpx_writer* w) {
static void update_coef_probs(VP9_COMP *cpi, vpx_writer *w) {
const TX_MODE tx_mode = cpi->common.tx_mode;
const TX_SIZE max_tx_size = tx_mode_to_biggest_tx_size[tx_mode];
TX_SIZE tx_size;
@@ -675,8 +666,7 @@ static void update_coef_probs(VP9_COMP *cpi, vpx_writer* w) {
(tx_size >= TX_16X16 && cpi->sf.tx_size_search_method == USE_TX_8X8)) {
vpx_write_bit(w, 0);
} else {
build_tree_distribution(cpi, tx_size, frame_branch_ct,
frame_coef_probs);
build_tree_distribution(cpi, tx_size, frame_branch_ct, frame_coef_probs);
update_coef_probs_common(w, cpi, tx_size, frame_branch_ct,
frame_coef_probs);
}
@@ -748,8 +738,7 @@ static void encode_segmentation(VP9_COMMON *cm, MACROBLOCKD *xd,
const struct segmentation *seg = &cm->seg;
vpx_wb_write_bit(wb, seg->enabled);
if (!seg->enabled)
return;
if (!seg->enabled) return;
// Segmentation map
vpx_wb_write_bit(wb, seg->update_map);
@@ -761,8 +750,7 @@ static void encode_segmentation(VP9_COMMON *cm, MACROBLOCKD *xd,
const int prob = seg->tree_probs[i];
const int update = prob != MAX_PROB;
vpx_wb_write_bit(wb, update);
if (update)
vpx_wb_write_literal(wb, prob, 8);
if (update) vpx_wb_write_literal(wb, prob, 8);
}
// Write out the chosen coding method.
@@ -772,8 +760,7 @@ static void encode_segmentation(VP9_COMMON *cm, MACROBLOCKD *xd,
const int prob = seg->pred_probs[i];
const int update = prob != MAX_PROB;
vpx_wb_write_bit(wb, update);
if (update)
vpx_wb_write_literal(wb, prob, 8);
if (update) vpx_wb_write_literal(wb, prob, 8);
}
}
}
@@ -817,7 +804,6 @@ static void encode_txfm_probs(VP9_COMMON *cm, vpx_writer *w,
unsigned int ct_16x16p[TX_SIZES - 2][2];
unsigned int ct_32x32p[TX_SIZES - 1][2];
for (i = 0; i < TX_SIZE_CONTEXTS; i++) {
tx_counts_to_branch_counts_8x8(counts->tx.p8x8[i], ct_8x8p);
for (j = 0; j < TX_SIZES - 3; j++)
@@ -879,16 +865,13 @@ static void write_tile_info(const VP9_COMMON *const cm,
// columns
ones = cm->log2_tile_cols - min_log2_tile_cols;
while (ones--)
vpx_wb_write_bit(wb, 1);
while (ones--) vpx_wb_write_bit(wb, 1);
if (cm->log2_tile_cols < max_log2_tile_cols)
vpx_wb_write_bit(wb, 0);
if (cm->log2_tile_cols < max_log2_tile_cols) vpx_wb_write_bit(wb, 0);
// rows
vpx_wb_write_bit(wb, cm->log2_tile_rows != 0);
if (cm->log2_tile_rows != 0)
vpx_wb_write_bit(wb, cm->log2_tile_rows != 1);
if (cm->log2_tile_rows != 0) vpx_wb_write_bit(wb, cm->log2_tile_rows != 1);
}
int vp9_get_refresh_mask(VP9_COMP *cpi) {
@@ -935,15 +918,15 @@ static size_t encode_tiles(VP9_COMP *cpi, uint8_t *data_ptr) {
TOKENEXTRA *tok = cpi->tile_tok[tile_row][tile_col];
tok_end = cpi->tile_tok[tile_row][tile_col] +
cpi->tok_count[tile_row][tile_col];
cpi->tok_count[tile_row][tile_col];
if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1)
vpx_start_encode(&residual_bc, data_ptr + total_size + 4);
else
vpx_start_encode(&residual_bc, data_ptr + total_size);
write_modes(cpi, &cpi->tile_data[tile_idx].tile_info,
&residual_bc, &tok, tok_end);
write_modes(cpi, &cpi->tile_data[tile_idx].tile_info, &residual_bc, &tok,
tok_end);
assert(tok == tok_end);
vpx_stop_encode(&residual_bc);
if (tile_col < tile_cols - 1 || tile_row < tile_rows - 1) {
@@ -961,8 +944,8 @@ static size_t encode_tiles(VP9_COMP *cpi, uint8_t *data_ptr) {
static void write_render_size(const VP9_COMMON *cm,
struct vpx_write_bit_buffer *wb) {
const int scaling_active = cm->width != cm->render_width ||
cm->height != cm->render_height;
const int scaling_active =
cm->width != cm->render_width || cm->height != cm->render_height;
vpx_wb_write_bit(wb, scaling_active);
if (scaling_active) {
vpx_wb_write_literal(wb, cm->render_width - 1, 16);
@@ -990,17 +973,17 @@ static void write_frame_size_with_refs(VP9_COMP *cpi,
// Set "found" to 0 for temporal svc and for spatial svc key frame
if (cpi->use_svc &&
((cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.rc_mode == VPX_CBR) ||
(cpi->svc.number_spatial_layers > 1 &&
cpi->svc.layer_context[cpi->svc.spatial_layer_id].is_key_frame) ||
(is_two_pass_svc(cpi) &&
cpi->svc.encode_empty_frame_state == ENCODING &&
cpi->svc.layer_context[0].frames_from_key_frame <
cpi->svc.number_temporal_layers + 1))) {
cpi->oxcf.rc_mode == VPX_CBR) ||
(cpi->svc.number_spatial_layers > 1 &&
cpi->svc.layer_context[cpi->svc.spatial_layer_id].is_key_frame) ||
(is_two_pass_svc(cpi) &&
cpi->svc.encode_empty_frame_state == ENCODING &&
cpi->svc.layer_context[0].frames_from_key_frame <
cpi->svc.number_temporal_layers + 1))) {
found = 0;
} else if (cfg != NULL) {
found = cm->width == cfg->y_crop_width &&
cm->height == cfg->y_crop_height;
found =
cm->width == cfg->y_crop_width && cm->height == cfg->y_crop_height;
}
vpx_wb_write_bit(wb, found);
if (found) {
@@ -1025,20 +1008,11 @@ static void write_sync_code(struct vpx_write_bit_buffer *wb) {
static void write_profile(BITSTREAM_PROFILE profile,
struct vpx_write_bit_buffer *wb) {
switch (profile) {
case PROFILE_0:
vpx_wb_write_literal(wb, 0, 2);
break;
case PROFILE_1:
vpx_wb_write_literal(wb, 2, 2);
break;
case PROFILE_2:
vpx_wb_write_literal(wb, 1, 2);
break;
case PROFILE_3:
vpx_wb_write_literal(wb, 6, 3);
break;
default:
assert(0);
case PROFILE_0: vpx_wb_write_literal(wb, 0, 2); break;
case PROFILE_1: vpx_wb_write_literal(wb, 2, 2); break;
case PROFILE_2: vpx_wb_write_literal(wb, 1, 2); break;
case PROFILE_3: vpx_wb_write_literal(wb, 6, 3); break;
default: assert(0);
}
}
@@ -1093,8 +1067,7 @@ static void write_uncompressed_header(VP9_COMP *cpi,
// will change to show_frame flag to 0, then add an one byte frame with
// show_existing_frame flag which tells the decoder which frame we want to
// show.
if (!cm->show_frame)
vpx_wb_write_bit(wb, cm->intra_only);
if (!cm->show_frame) vpx_wb_write_bit(wb, cm->intra_only);
if (!cm->error_resilient_mode)
vpx_wb_write_literal(wb, cm->reset_frame_context, 2);
@@ -1222,7 +1195,7 @@ static size_t write_compressed_header(VP9_COMP *cpi, uint8_t *data) {
void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, size_t *size) {
uint8_t *data = dest;
size_t first_part_size, uncompressed_hdr_size;
struct vpx_write_bit_buffer wb = {data, 0};
struct vpx_write_bit_buffer wb = { data, 0 };
struct vpx_write_bit_buffer saved_wb;
write_uncompressed_header(cpi, &wb);

8
vp9/encoder/vp9_bitstream.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_BITSTREAM_H_
#define VP9_ENCODER_VP9_BITSTREAM_H_
@@ -25,10 +24,9 @@ void vp9_pack_bitstream(VP9_COMP *cpi, uint8_t *dest, size_t *size);
static INLINE int vp9_preserve_existing_gf(VP9_COMP *cpi) {
return !cpi->multi_arf_allowed && cpi->refresh_golden_frame &&
cpi->rc.is_src_frame_alt_ref &&
(!cpi->use_svc || // Add spatial svc base layer case here
(is_two_pass_svc(cpi) &&
cpi->svc.spatial_layer_id == 0 &&
cpi->svc.layer_context[0].gold_ref_idx >=0 &&
(!cpi->use_svc || // Add spatial svc base layer case here
(is_two_pass_svc(cpi) && cpi->svc.spatial_layer_id == 0 &&
cpi->svc.layer_context[0].gold_ref_idx >= 0 &&
cpi->oxcf.ss_enable_auto_arf[0]));
}

10
vp9/encoder/vp9_block.h
View File

@@ -77,8 +77,8 @@ struct macroblock {
int rddiv;
int rdmult;
int mb_energy;
int * m_search_count_ptr;
int * ex_search_count_ptr;
int *m_search_count_ptr;
int *ex_search_count_ptr;
// These are set to their default values at the beginning, and then adjusted
// further in the encoding process.
@@ -130,9 +130,9 @@ struct macroblock {
// skip forward transform and quantization
uint8_t skip_txfm[MAX_MB_PLANE << 2];
#define SKIP_TXFM_NONE 0
#define SKIP_TXFM_AC_DC 1
#define SKIP_TXFM_AC_ONLY 2
#define SKIP_TXFM_NONE 0
#define SKIP_TXFM_AC_DC 1
#define SKIP_TXFM_AC_ONLY 2
int64_t bsse[MAX_MB_PLANE << 2];

20
vp9/encoder/vp9_blockiness.c
View File

@@ -63,9 +63,9 @@ static int blockiness_vertical(const uint8_t *s, int sp, const uint8_t *r,
s_blockiness += horizontal_filter(s);
r_blockiness += horizontal_filter(r);
sum_0 += s[0];
sum_sq_0 += s[0]*s[0];
sum_sq_0 += s[0] * s[0];
sum_1 += s[-1];
sum_sq_1 += s[-1]*s[-1];
sum_sq_1 += s[-1] * s[-1];
}
var_0 = variance(sum_0, sum_sq_0, size);
var_1 = variance(sum_1, sum_sq_1, size);
@@ -113,19 +113,19 @@ static int blockiness_horizontal(const uint8_t *s, int sp, const uint8_t *r,
// This function returns the blockiness for the entire frame currently by
// looking at all borders in steps of 4.
double vp9_get_blockiness(const uint8_t *img1, int img1_pitch,
const uint8_t *img2, int img2_pitch,
int width, int height) {
const uint8_t *img2, int img2_pitch, int width,
int height) {
double blockiness = 0;
int i, j;
vpx_clear_system_state();
for (i = 0; i < height; i += 4, img1 += img1_pitch * 4,
img2 += img2_pitch * 4) {
for (i = 0; i < height;
i += 4, img1 += img1_pitch * 4, img2 += img2_pitch * 4) {
for (j = 0; j < width; j += 4) {
if (i > 0 && i < height && j > 0 && j < width) {
blockiness += blockiness_vertical(img1 + j, img1_pitch,
img2 + j, img2_pitch, 4);
blockiness += blockiness_horizontal(img1 + j, img1_pitch,
img2 + j, img2_pitch, 4);
blockiness +=
blockiness_vertical(img1 + j, img1_pitch, img2 + j, img2_pitch, 4);
blockiness += blockiness_horizontal(img1 + j, img1_pitch, img2 + j,
img2_pitch, 4);
}
}
}

45
vp9/encoder/vp9_context_tree.c
View File

@@ -12,10 +12,7 @@
#include "vp9/encoder/vp9_encoder.h"
static const BLOCK_SIZE square[] = {
BLOCK_8X8,
BLOCK_16X16,
BLOCK_32X32,
BLOCK_64X64,
BLOCK_8X8, BLOCK_16X16, BLOCK_32X32, BLOCK_64X64,
};
static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk,
@@ -25,8 +22,7 @@ static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk,
int i, k;
ctx->num_4x4_blk = num_blk;
CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
vpx_calloc(num_blk, sizeof(uint8_t)));
CHECK_MEM_ERROR(cm, ctx->zcoeff_blk, vpx_calloc(num_blk, sizeof(uint8_t)));
for (i = 0; i < MAX_MB_PLANE; ++i) {
for (k = 0; k < 3; ++k) {
CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
@@ -37,10 +33,10 @@ static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk,
vpx_memalign(32, num_pix * sizeof(*ctx->dqcoeff[i][k])));
CHECK_MEM_ERROR(cm, ctx->eobs[i][k],
vpx_memalign(32, num_blk * sizeof(*ctx->eobs[i][k])));
ctx->coeff_pbuf[i][k] = ctx->coeff[i][k];
ctx->qcoeff_pbuf[i][k] = ctx->qcoeff[i][k];
ctx->coeff_pbuf[i][k] = ctx->coeff[i][k];
ctx->qcoeff_pbuf[i][k] = ctx->qcoeff[i][k];
ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k];
ctx->eobs_pbuf[i][k] = ctx->eobs[i][k];
ctx->eobs_pbuf[i][k] = ctx->eobs[i][k];
}
}
}
@@ -66,12 +62,12 @@ static void free_mode_context(PICK_MODE_CONTEXT *ctx) {
static void alloc_tree_contexts(VP9_COMMON *cm, PC_TREE *tree,
int num_4x4_blk) {
alloc_mode_context(cm, num_4x4_blk, &tree->none);
alloc_mode_context(cm, num_4x4_blk/2, &tree->horizontal[0]);
alloc_mode_context(cm, num_4x4_blk/2, &tree->vertical[0]);
alloc_mode_context(cm, num_4x4_blk / 2, &tree->horizontal[0]);
alloc_mode_context(cm, num_4x4_blk / 2, &tree->vertical[0]);
if (num_4x4_blk > 4) {
alloc_mode_context(cm, num_4x4_blk/2, &tree->horizontal[1]);
alloc_mode_context(cm, num_4x4_blk/2, &tree->vertical[1]);
alloc_mode_context(cm, num_4x4_blk / 2, &tree->horizontal[1]);
alloc_mode_context(cm, num_4x4_blk / 2, &tree->vertical[1]);
} else {
memset(&tree->horizontal[1], 0, sizeof(tree->horizontal[1]));
memset(&tree->vertical[1], 0, sizeof(tree->vertical[1]));
@@ -101,19 +97,18 @@ void vp9_setup_pc_tree(VP9_COMMON *cm, ThreadData *td) {
int nodes;
vpx_free(td->leaf_tree);
CHECK_MEM_ERROR(cm, td->leaf_tree, vpx_calloc(leaf_nodes,
sizeof(*td->leaf_tree)));
CHECK_MEM_ERROR(cm, td->leaf_tree,
vpx_calloc(leaf_nodes, sizeof(*td->leaf_tree)));
vpx_free(td->pc_tree);
CHECK_MEM_ERROR(cm, td->pc_tree, vpx_calloc(tree_nodes,
sizeof(*td->pc_tree)));
CHECK_MEM_ERROR(cm, td->pc_tree,
vpx_calloc(tree_nodes, sizeof(*td->pc_tree)));
this_pc = &td->pc_tree[0];
this_leaf = &td->leaf_tree[0];
// 4x4 blocks smaller than 8x8 but in the same 8x8 block share the same
// context so we only need to allocate 1 for each 8x8 block.
for (i = 0; i < leaf_nodes; ++i)
alloc_mode_context(cm, 1, &td->leaf_tree[i]);
for (i = 0; i < leaf_nodes; ++i) alloc_mode_context(cm, 1, &td->leaf_tree[i]);
// Sets up all the leaf nodes in the tree.
for (pc_tree_index = 0; pc_tree_index < leaf_nodes; ++pc_tree_index) {
@@ -121,8 +116,7 @@ void vp9_setup_pc_tree(VP9_COMMON *cm, ThreadData *td) {
tree->block_size = square[0];
alloc_tree_contexts(cm, tree, 4);
tree->leaf_split[0] = this_leaf++;
for (j = 1; j < 4; j++)
tree->leaf_split[j] = tree->leaf_split[0];
for (j = 1; j < 4; j++) tree->leaf_split[j] = tree->leaf_split[0];
}
// Each node has 4 leaf nodes, fill each block_size level of the tree
@@ -132,8 +126,7 @@ void vp9_setup_pc_tree(VP9_COMMON *cm, ThreadData *td) {
PC_TREE *const tree = &td->pc_tree[pc_tree_index];
alloc_tree_contexts(cm, tree, 4 << (2 * square_index));
tree->block_size = square[square_index];
for (j = 0; j < 4; j++)
tree->split[j] = this_pc++;
for (j = 0; j < 4; j++) tree->split[j] = this_pc++;
++pc_tree_index;
}
++square_index;
@@ -147,12 +140,10 @@ void vp9_free_pc_tree(ThreadData *td) {
int i;
// Set up all 4x4 mode contexts
for (i = 0; i < 64; ++i)
free_mode_context(&td->leaf_tree[i]);
for (i = 0; i < 64; ++i) free_mode_context(&td->leaf_tree[i]);
// Sets up all the leaf nodes in the tree.
for (i = 0; i < tree_nodes; ++i)
free_tree_contexts(&td->pc_tree[i]);
for (i = 0; i < tree_nodes; ++i) free_tree_contexts(&td->pc_tree[i]);
vpx_free(td->pc_tree);
td->pc_tree = NULL;

47
vp9/encoder/vp9_cost.c
View File

@@ -14,31 +14,30 @@
/* round(-log2(i/256.) * (1 << VP9_PROB_COST_SHIFT))
Begins with a bogus entry for simpler addressing. */
const uint16_t vp9_prob_cost[256] = {
4096, 4096, 3584, 3284, 3072, 2907, 2772, 2659, 2560, 2473, 2395, 2325,
2260, 2201, 2147, 2096, 2048, 2003, 1961, 1921, 1883, 1847, 1813, 1780,
1748, 1718, 1689, 1661, 1635, 1609, 1584, 1559, 1536, 1513, 1491, 1470,
1449, 1429, 1409, 1390, 1371, 1353, 1335, 1318, 1301, 1284, 1268, 1252,
1236, 1221, 1206, 1192, 1177, 1163, 1149, 1136, 1123, 1110, 1097, 1084,
1072, 1059, 1047, 1036, 1024, 1013, 1001, 990, 979, 968, 958, 947,
937, 927, 917, 907, 897, 887, 878, 868, 859, 850, 841, 832,
823, 814, 806, 797, 789, 780, 772, 764, 756, 748, 740, 732,
724, 717, 709, 702, 694, 687, 680, 673, 665, 658, 651, 644,
637, 631, 624, 617, 611, 604, 598, 591, 585, 578, 572, 566,
560, 554, 547, 541, 535, 530, 524, 518, 512, 506, 501, 495,
489, 484, 478, 473, 467, 462, 456, 451, 446, 441, 435, 430,
425, 420, 415, 410, 405, 400, 395, 390, 385, 380, 375, 371,
366, 361, 356, 352, 347, 343, 338, 333, 329, 324, 320, 316,
311, 307, 302, 298, 294, 289, 285, 281, 277, 273, 268, 264,
260, 256, 252, 248, 244, 240, 236, 232, 228, 224, 220, 216,
212, 209, 205, 201, 197, 194, 190, 186, 182, 179, 175, 171,
168, 164, 161, 157, 153, 150, 146, 143, 139, 136, 132, 129,
125, 122, 119, 115, 112, 109, 105, 102, 99, 95, 92, 89,
86, 82, 79, 76, 73, 70, 66, 63, 60, 57, 54, 51,
48, 45, 42, 38, 35, 32, 29, 26, 23, 20, 18, 15,
12, 9, 6, 3};
4096, 4096, 3584, 3284, 3072, 2907, 2772, 2659, 2560, 2473, 2395, 2325, 2260,
2201, 2147, 2096, 2048, 2003, 1961, 1921, 1883, 1847, 1813, 1780, 1748, 1718,
1689, 1661, 1635, 1609, 1584, 1559, 1536, 1513, 1491, 1470, 1449, 1429, 1409,
1390, 1371, 1353, 1335, 1318, 1301, 1284, 1268, 1252, 1236, 1221, 1206, 1192,
1177, 1163, 1149, 1136, 1123, 1110, 1097, 1084, 1072, 1059, 1047, 1036, 1024,
1013, 1001, 990, 979, 968, 958, 947, 937, 927, 917, 907, 897, 887,
878, 868, 859, 850, 841, 832, 823, 814, 806, 797, 789, 780, 772,
764, 756, 748, 740, 732, 724, 717, 709, 702, 694, 687, 680, 673,
665, 658, 651, 644, 637, 631, 624, 617, 611, 604, 598, 591, 585,
578, 572, 566, 560, 554, 547, 541, 535, 530, 524, 518, 512, 506,
501, 495, 489, 484, 478, 473, 467, 462, 456, 451, 446, 441, 435,
430, 425, 420, 415, 410, 405, 400, 395, 390, 385, 380, 375, 371,
366, 361, 356, 352, 347, 343, 338, 333, 329, 324, 320, 316, 311,
307, 302, 298, 294, 289, 285, 281, 277, 273, 268, 264, 260, 256,
252, 248, 244, 240, 236, 232, 228, 224, 220, 216, 212, 209, 205,
201, 197, 194, 190, 186, 182, 179, 175, 171, 168, 164, 161, 157,
153, 150, 146, 143, 139, 136, 132, 129, 125, 122, 119, 115, 112,
109, 105, 102, 99, 95, 92, 89, 86, 82, 79, 76, 73, 70,
66, 63, 60, 57, 54, 51, 48, 45, 42, 38, 35, 32, 29,
26, 23, 20, 18, 15, 12, 9, 6, 3
};
static void cost(int *costs, vpx_tree tree, const vpx_prob *probs,
int i, int c) {
static void cost(int *costs, vpx_tree tree, const vpx_prob *probs, int i,
int c) {
const vpx_prob prob = probs[i / 2];
int b;

7
vp9/encoder/vp9_cost.h
View File

@@ -27,16 +27,15 @@ extern const uint16_t vp9_prob_cost[256];
#define vp9_cost_one(prob) vp9_cost_zero(256 - (prob))
#define vp9_cost_bit(prob, bit) vp9_cost_zero((bit) ? 256 - (prob) \
: (prob))
#define vp9_cost_bit(prob, bit) vp9_cost_zero((bit) ? 256 - (prob) : (prob))
static INLINE unsigned int cost_branch256(const unsigned int ct[2],
vpx_prob p) {
return ct[0] * vp9_cost_zero(p) + ct[1] * vp9_cost_one(p);
}
static INLINE int treed_cost(vpx_tree tree, const vpx_prob *probs,
int bits, int len) {
static INLINE int treed_cost(vpx_tree tree, const vpx_prob *probs, int bits,
int len) {
int cost = 0;
vpx_tree_index i = 0;

217
vp9/encoder/vp9_dct.c
View File

@@ -61,8 +61,8 @@ static void fdct8(const tran_low_t *input, tran_low_t *output) {
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
output[0] = (tran_low_t)fdct_round_shift(t0);
output[2] = (tran_low_t)fdct_round_shift(t2);
output[4] = (tran_low_t)fdct_round_shift(t1);
@@ -81,10 +81,10 @@ static void fdct8(const tran_low_t *input, tran_low_t *output) {
x3 = s7 + t3;
// Stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
output[1] = (tran_low_t)fdct_round_shift(t0);
output[3] = (tran_low_t)fdct_round_shift(t2);
output[5] = (tran_low_t)fdct_round_shift(t1);
@@ -105,11 +105,11 @@ static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
input[3] = in[3] + in[12];
input[4] = in[4] + in[11];
input[5] = in[5] + in[10];
input[6] = in[6] + in[ 9];
input[7] = in[7] + in[ 8];
input[6] = in[6] + in[9];
input[7] = in[7] + in[8];
step1[0] = in[7] - in[ 8];
step1[1] = in[6] - in[ 9];
step1[0] = in[7] - in[8];
step1[1] = in[6] - in[9];
step1[2] = in[5] - in[10];
step1[3] = in[4] - in[11];
step1[4] = in[3] - in[12];
@@ -140,7 +140,7 @@ static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
out[0] = (tran_low_t)fdct_round_shift(t0);
out[4] = (tran_low_t)fdct_round_shift(t2);
@@ -160,10 +160,10 @@ static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
x3 = s7 + t3;
// Stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
out[2] = (tran_low_t)fdct_round_shift(t0);
out[6] = (tran_low_t)fdct_round_shift(t2);
out[10] = (tran_low_t)fdct_round_shift(t1);
@@ -191,12 +191,12 @@ static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
step3[7] = step1[7] + step2[4];
// step 4
temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
temp2 = step3[2] * cospi_24_64 + step3[5] * cospi_8_64;
temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
temp2 = step3[2] * cospi_24_64 + step3[5] * cospi_8_64;
step2[1] = fdct_round_shift(temp1);
step2[2] = fdct_round_shift(temp2);
temp1 = step3[2] * cospi_8_64 - step3[5] * cospi_24_64;
temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
step2[5] = fdct_round_shift(temp1);
step2[6] = fdct_round_shift(temp2);
@@ -211,23 +211,23 @@ static void fdct16(const tran_low_t in[16], tran_low_t out[16]) {
step1[7] = step3[7] + step2[6];
// step 6
temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
out[1] = (tran_low_t)fdct_round_shift(temp1);
out[9] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
out[5] = (tran_low_t)fdct_round_shift(temp1);
out[13] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
out[3] = (tran_low_t)fdct_round_shift(temp1);
out[11] = (tran_low_t)fdct_round_shift(temp2);
temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
out[7] = (tran_low_t)fdct_round_shift(temp1);
out[15] = (tran_low_t)fdct_round_shift(temp2);
}
@@ -285,14 +285,14 @@ static void fadst8(const tran_low_t *input, tran_low_t *output) {
tran_high_t x7 = input[6];
// stage 1
s0 = cospi_2_64 * x0 + cospi_30_64 * x1;
s1 = cospi_30_64 * x0 - cospi_2_64 * x1;
s0 = cospi_2_64 * x0 + cospi_30_64 * x1;
s1 = cospi_30_64 * x0 - cospi_2_64 * x1;
s2 = cospi_10_64 * x2 + cospi_22_64 * x3;
s3 = cospi_22_64 * x2 - cospi_10_64 * x3;
s4 = cospi_18_64 * x4 + cospi_14_64 * x5;
s5 = cospi_14_64 * x4 - cospi_18_64 * x5;
s6 = cospi_26_64 * x6 + cospi_6_64 * x7;
s7 = cospi_6_64 * x6 - cospi_26_64 * x7;
s6 = cospi_26_64 * x6 + cospi_6_64 * x7;
s7 = cospi_6_64 * x6 - cospi_26_64 * x7;
x0 = fdct_round_shift(s0 + s4);
x1 = fdct_round_shift(s1 + s5);
@@ -308,10 +308,10 @@ static void fadst8(const tran_low_t *input, tran_low_t *output) {
s1 = x1;
s2 = x2;
s3 = x3;
s4 = cospi_8_64 * x4 + cospi_24_64 * x5;
s5 = cospi_24_64 * x4 - cospi_8_64 * x5;
s6 = - cospi_24_64 * x6 + cospi_8_64 * x7;
s7 = cospi_8_64 * x6 + cospi_24_64 * x7;
s4 = cospi_8_64 * x4 + cospi_24_64 * x5;
s5 = cospi_24_64 * x4 - cospi_8_64 * x5;
s6 = -cospi_24_64 * x6 + cospi_8_64 * x7;
s7 = cospi_8_64 * x6 + cospi_24_64 * x7;
x0 = s0 + s2;
x1 = s1 + s3;
@@ -365,11 +365,11 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
tran_high_t x15 = input[14];
// stage 1
s0 = x0 * cospi_1_64 + x1 * cospi_31_64;
s0 = x0 * cospi_1_64 + x1 * cospi_31_64;
s1 = x0 * cospi_31_64 - x1 * cospi_1_64;
s2 = x2 * cospi_5_64 + x3 * cospi_27_64;
s2 = x2 * cospi_5_64 + x3 * cospi_27_64;
s3 = x2 * cospi_27_64 - x3 * cospi_5_64;
s4 = x4 * cospi_9_64 + x5 * cospi_23_64;
s4 = x4 * cospi_9_64 + x5 * cospi_23_64;
s5 = x4 * cospi_23_64 - x5 * cospi_9_64;
s6 = x6 * cospi_13_64 + x7 * cospi_19_64;
s7 = x6 * cospi_19_64 - x7 * cospi_13_64;
@@ -378,9 +378,9 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
s10 = x10 * cospi_21_64 + x11 * cospi_11_64;
s11 = x10 * cospi_11_64 - x11 * cospi_21_64;
s12 = x12 * cospi_25_64 + x13 * cospi_7_64;
s13 = x12 * cospi_7_64 - x13 * cospi_25_64;
s13 = x12 * cospi_7_64 - x13 * cospi_25_64;
s14 = x14 * cospi_29_64 + x15 * cospi_3_64;
s15 = x14 * cospi_3_64 - x15 * cospi_29_64;
s15 = x14 * cospi_3_64 - x15 * cospi_29_64;
x0 = fdct_round_shift(s0 + s8);
x1 = fdct_round_shift(s1 + s9);
@@ -390,8 +390,8 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
x5 = fdct_round_shift(s5 + s13);
x6 = fdct_round_shift(s6 + s14);
x7 = fdct_round_shift(s7 + s15);
x8 = fdct_round_shift(s0 - s8);
x9 = fdct_round_shift(s1 - s9);
x8 = fdct_round_shift(s0 - s8);
x9 = fdct_round_shift(s1 - s9);
x10 = fdct_round_shift(s2 - s10);
x11 = fdct_round_shift(s3 - s11);
x12 = fdct_round_shift(s4 - s12);
@@ -408,14 +408,14 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
s5 = x5;
s6 = x6;
s7 = x7;
s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
s12 = - x12 * cospi_28_64 + x13 * cospi_4_64;
s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
s14 = - x14 * cospi_12_64 + x15 * cospi_20_64;
s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
s8 = x8 * cospi_4_64 + x9 * cospi_28_64;
s9 = x8 * cospi_28_64 - x9 * cospi_4_64;
s10 = x10 * cospi_20_64 + x11 * cospi_12_64;
s11 = x10 * cospi_12_64 - x11 * cospi_20_64;
s12 = -x12 * cospi_28_64 + x13 * cospi_4_64;
s13 = x12 * cospi_4_64 + x13 * cospi_28_64;
s14 = -x14 * cospi_12_64 + x15 * cospi_20_64;
s15 = x14 * cospi_20_64 + x15 * cospi_12_64;
x0 = s0 + s4;
x1 = s1 + s5;
@@ -439,18 +439,18 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
s1 = x1;
s2 = x2;
s3 = x3;
s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
s4 = x4 * cospi_8_64 + x5 * cospi_24_64;
s5 = x4 * cospi_24_64 - x5 * cospi_8_64;
s6 = - x6 * cospi_24_64 + x7 * cospi_8_64;
s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
s6 = -x6 * cospi_24_64 + x7 * cospi_8_64;
s7 = x6 * cospi_8_64 + x7 * cospi_24_64;
s8 = x8;
s9 = x9;
s10 = x10;
s11 = x11;
s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
s12 = x12 * cospi_8_64 + x13 * cospi_24_64;
s13 = x12 * cospi_24_64 - x13 * cospi_8_64;
s14 = - x14 * cospi_24_64 + x15 * cospi_8_64;
s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
s14 = -x14 * cospi_24_64 + x15 * cospi_8_64;
s15 = x14 * cospi_8_64 + x15 * cospi_24_64;
x0 = s0 + s2;
x1 = s1 + s3;
@@ -470,13 +470,13 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
x15 = fdct_round_shift(s13 - s15);
// stage 4
s2 = (- cospi_16_64) * (x2 + x3);
s2 = (-cospi_16_64) * (x2 + x3);
s3 = cospi_16_64 * (x2 - x3);
s6 = cospi_16_64 * (x6 + x7);
s7 = cospi_16_64 * (- x6 + x7);
s7 = cospi_16_64 * (-x6 + x7);
s10 = cospi_16_64 * (x10 + x11);
s11 = cospi_16_64 * (- x10 + x11);
s14 = (- cospi_16_64) * (x14 + x15);
s11 = cospi_16_64 * (-x10 + x11);
s14 = (-cospi_16_64) * (x14 + x15);
s15 = cospi_16_64 * (x14 - x15);
x2 = fdct_round_shift(s2);
@@ -507,28 +507,28 @@ static void fadst16(const tran_low_t *input, tran_low_t *output) {
}
static const transform_2d FHT_4[] = {
{ fdct4, fdct4 }, // DCT_DCT = 0
{ fadst4, fdct4 }, // ADST_DCT = 1
{ fdct4, fadst4 }, // DCT_ADST = 2
{ fadst4, fadst4 } // ADST_ADST = 3
{ fdct4, fdct4 }, // DCT_DCT = 0
{ fadst4, fdct4 }, // ADST_DCT = 1
{ fdct4, fadst4 }, // DCT_ADST = 2
{ fadst4, fadst4 } // ADST_ADST = 3
};
static const transform_2d FHT_8[] = {
{ fdct8, fdct8 }, // DCT_DCT = 0
{ fadst8, fdct8 }, // ADST_DCT = 1
{ fdct8, fadst8 }, // DCT_ADST = 2
{ fadst8, fadst8 } // ADST_ADST = 3
{ fdct8, fdct8 }, // DCT_DCT = 0
{ fadst8, fdct8 }, // ADST_DCT = 1
{ fdct8, fadst8 }, // DCT_ADST = 2
{ fadst8, fadst8 } // ADST_ADST = 3
};
static const transform_2d FHT_16[] = {
{ fdct16, fdct16 }, // DCT_DCT = 0
{ fadst16, fdct16 }, // ADST_DCT = 1
{ fdct16, fadst16 }, // DCT_ADST = 2
{ fadst16, fadst16 } // ADST_ADST = 3
{ fdct16, fdct16 }, // DCT_DCT = 0
{ fadst16, fdct16 }, // ADST_DCT = 1
{ fdct16, fadst16 }, // DCT_ADST = 2
{ fadst16, fadst16 } // ADST_ADST = 3
};
void vp9_fht4x4_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_fht4x4_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
if (tx_type == DCT_DCT) {
vpx_fdct4x4_c(input, output, stride);
} else {
@@ -539,36 +539,29 @@ void vp9_fht4x4_c(const int16_t *input, tran_low_t *output,
// Columns
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j)
temp_in[j] = input[j * stride + i] * 16;
if (i == 0 && temp_in[0])
temp_in[0] += 1;
for (j = 0; j < 4; ++j) temp_in[j] = input[j * stride + i] * 16;
if (i == 0 && temp_in[0]) temp_in[0] += 1;
ht.cols(temp_in, temp_out);
for (j = 0; j < 4; ++j)
out[j * 4 + i] = temp_out[j];
for (j = 0; j < 4; ++j) out[j * 4 + i] = temp_out[j];
}
// Rows
for (i = 0; i < 4; ++i) {
for (j = 0; j < 4; ++j)
temp_in[j] = out[j + i * 4];
for (j = 0; j < 4; ++j) temp_in[j] = out[j + i * 4];
ht.rows(temp_in, temp_out);
for (j = 0; j < 4; ++j)
output[j + i * 4] = (temp_out[j] + 1) >> 2;
for (j = 0; j < 4; ++j) output[j + i * 4] = (temp_out[j] + 1) >> 2;
}
}
}
void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block,
const int16_t *zbin_ptr, const int16_t *round_ptr,
const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan,
const int16_t *iscan) {
int eob = -1;
int i, j;
@@ -600,8 +593,8 @@ void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
x3 = s0 - s3;
t0 = (x0 + x1) * cospi_16_64;
t1 = (x0 - x1) * cospi_16_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
output[0 * 8] = (tran_low_t)fdct_round_shift(t0);
output[2 * 8] = (tran_low_t)fdct_round_shift(t2);
output[4 * 8] = (tran_low_t)fdct_round_shift(t1);
@@ -620,10 +613,10 @@ void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
x3 = s7 + t3;
// Stage 4
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
output[1 * 8] = (tran_low_t)fdct_round_shift(t0);
output[3 * 8] = (tran_low_t)fdct_round_shift(t2);
output[5 * 8] = (tran_low_t)fdct_round_shift(t1);
@@ -636,8 +629,7 @@ void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
// Rows
for (i = 0; i < 8; ++i) {
fdct8(&intermediate[i * 8], &coeff_ptr[i * 8]);
for (j = 0; j < 8; ++j)
coeff_ptr[j + i * 8] /= 2;
for (j = 0; j < 8; ++j) coeff_ptr[j + i * 8] /= 2;
}
// TODO(jingning) Decide the need of these arguments after the
@@ -664,15 +656,14 @@ void vp9_fdct8x8_quant_c(const int16_t *input, int stride,
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
if (tmp)
eob = i;
if (tmp) eob = i;
}
}
*eob_ptr = eob + 1;
}
void vp9_fht8x8_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_fht8x8_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
if (tx_type == DCT_DCT) {
vpx_fdct8x8_c(input, output, stride);
} else {
@@ -683,17 +674,14 @@ void vp9_fht8x8_c(const int16_t *input, tran_low_t *output,
// Columns
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j)
temp_in[j] = input[j * stride + i] * 4;
for (j = 0; j < 8; ++j) temp_in[j] = input[j * stride + i] * 4;
ht.cols(temp_in, temp_out);
for (j = 0; j < 8; ++j)
out[j * 8 + i] = temp_out[j];
for (j = 0; j < 8; ++j) out[j * 8 + i] = temp_out[j];
}
// Rows
for (i = 0; i < 8; ++i) {
for (j = 0; j < 8; ++j)
temp_in[j] = out[j + i * 8];
for (j = 0; j < 8; ++j) temp_in[j] = out[j + i * 8];
ht.rows(temp_in, temp_out);
for (j = 0; j < 8; ++j)
output[j + i * 8] = (temp_out[j] + (temp_out[j] < 0)) >> 1;
@@ -757,8 +745,8 @@ void vp9_fwht4x4_c(const int16_t *input, tran_low_t *output, int stride) {
}
}
void vp9_fht16x16_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_fht16x16_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
if (tx_type == DCT_DCT) {
vpx_fdct16x16_c(input, output, stride);
} else {
@@ -769,8 +757,7 @@ void vp9_fht16x16_c(const int16_t *input, tran_low_t *output,
// Columns
for (i = 0; i < 16; ++i) {
for (j = 0; j < 16; ++j)
temp_in[j] = input[j * stride + i] * 4;
for (j = 0; j < 16; ++j) temp_in[j] = input[j * stride + i] * 4;
ht.cols(temp_in, temp_out);
for (j = 0; j < 16; ++j)
out[j * 16 + i] = (temp_out[j] + 1 + (temp_out[j] < 0)) >> 2;
@@ -778,23 +765,21 @@ void vp9_fht16x16_c(const int16_t *input, tran_low_t *output,
// Rows
for (i = 0; i < 16; ++i) {
for (j = 0; j < 16; ++j)
temp_in[j] = out[j + i * 16];
for (j = 0; j < 16; ++j) temp_in[j] = out[j + i * 16];
ht.rows(temp_in, temp_out);
for (j = 0; j < 16; ++j)
output[j + i * 16] = temp_out[j];
for (j = 0; j < 16; ++j) output[j + i * 16] = temp_out[j];
}
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_fht4x4_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_highbd_fht4x4_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
vp9_fht4x4_c(input, output, stride, tx_type);
}
void vp9_highbd_fht8x8_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_highbd_fht8x8_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
vp9_fht8x8_c(input, output, stride, tx_type);
}
@@ -803,8 +788,8 @@ void vp9_highbd_fwht4x4_c(const int16_t *input, tran_low_t *output,
vp9_fwht4x4_c(input, output, stride);
}
void vp9_highbd_fht16x16_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_highbd_fht16x16_c(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
vp9_fht16x16_c(input, output, stride, tx_type);
}
#endif // CONFIG_VP9_HIGHBITDEPTH

196
vp9/encoder/vp9_denoiser.c
View File

@@ -48,8 +48,7 @@ static unsigned int sse_thresh(BLOCK_SIZE bs, int increase_denoising) {
static int sse_diff_thresh(BLOCK_SIZE bs, int increase_denoising,
int motion_magnitude) {
if (motion_magnitude >
noise_motion_thresh(bs, increase_denoising)) {
if (motion_magnitude > noise_motion_thresh(bs, increase_denoising)) {
if (increase_denoising)
return (1 << num_pels_log2_lookup[bs]) << 2;
else
@@ -67,18 +66,15 @@ static int total_adj_weak_thresh(BLOCK_SIZE bs, int increase_denoising) {
// we might need to update the code for calculating 'total_adj' in
// case the C code is not bit-exact with corresponding sse2 code.
int vp9_denoiser_filter_c(const uint8_t *sig, int sig_stride,
const uint8_t *mc_avg,
int mc_avg_stride,
uint8_t *avg, int avg_stride,
int increase_denoising,
BLOCK_SIZE bs,
int motion_magnitude) {
const uint8_t *mc_avg, int mc_avg_stride,
uint8_t *avg, int avg_stride, int increase_denoising,
BLOCK_SIZE bs, int motion_magnitude) {
int r, c;
const uint8_t *sig_start = sig;
const uint8_t *mc_avg_start = mc_avg;
uint8_t *avg_start = avg;
int diff, adj, absdiff, delta;
int adj_val[] = {3, 4, 6};
int adj_val[] = { 3, 4, 6 };
int total_adj = 0;
int shift_inc = 1;
@@ -105,15 +101,19 @@ int vp9_denoiser_filter_c(const uint8_t *sig, int sig_stride,
total_adj += diff;
} else {
switch (absdiff) {
case 4: case 5: case 6: case 7:
adj = adj_val[0];
break;
case 8: case 9: case 10: case 11:
case 12: case 13: case 14: case 15:
adj = adj_val[1];
break;
default:
adj = adj_val[2];
case 4:
case 5:
case 6:
case 7: adj = adj_val[0]; break;
case 8:
case 9:
case 10:
case 11:
case 12:
case 13:
case 14:
case 15: adj = adj_val[1]; break;
default: adj = adj_val[2];
}
if (diff > 0) {
avg[c] = VPXMIN(UINT8_MAX, sig[c] + adj);
@@ -135,14 +135,15 @@ int vp9_denoiser_filter_c(const uint8_t *sig, int sig_stride,
}
// Otherwise, we try to dampen the filter if the delta is not too high.
delta = ((abs(total_adj) - total_adj_strong_thresh(bs, increase_denoising))
>> num_pels_log2_lookup[bs]) + 1;
delta = ((abs(total_adj) - total_adj_strong_thresh(bs, increase_denoising)) >>
num_pels_log2_lookup[bs]) +
1;
if (delta >= delta_thresh(bs, increase_denoising)) {
return COPY_BLOCK;
}
mc_avg = mc_avg_start;
mc_avg = mc_avg_start;
avg = avg_start;
sig = sig_start;
for (r = 0; r < (4 << b_height_log2_lookup[bs]); ++r) {
@@ -178,22 +179,15 @@ int vp9_denoiser_filter_c(const uint8_t *sig, int sig_stride,
return COPY_BLOCK;
}
static uint8_t *block_start(uint8_t *framebuf, int stride,
int mi_row, int mi_col) {
return framebuf + (stride * mi_row << 3) + (mi_col << 3);
static uint8_t *block_start(uint8_t *framebuf, int stride, int mi_row,
int mi_col) {
return framebuf + (stride * mi_row << 3) + (mi_col << 3);
}
static VP9_DENOISER_DECISION perform_motion_compensation(VP9_DENOISER *denoiser,
MACROBLOCK *mb,
BLOCK_SIZE bs,
int increase_denoising,
int mi_row,
int mi_col,
PICK_MODE_CONTEXT *ctx,
int motion_magnitude,
int is_skin,
int *zeromv_filter,
int consec_zeromv) {
static VP9_DENOISER_DECISION perform_motion_compensation(
VP9_DENOISER *denoiser, MACROBLOCK *mb, BLOCK_SIZE bs,
int increase_denoising, int mi_row, int mi_col, PICK_MODE_CONTEXT *ctx,
int motion_magnitude, int is_skin, int *zeromv_filter, int consec_zeromv) {
int sse_diff = ctx->zeromv_sse - ctx->newmv_sse;
MV_REFERENCE_FRAME frame;
MACROBLOCKD *filter_mbd = &mb->e_mbd;
@@ -206,20 +200,18 @@ static VP9_DENOISER_DECISION perform_motion_compensation(VP9_DENOISER *denoiser,
frame = ctx->best_reference_frame;
saved_mi = *mi;
if (is_skin && (motion_magnitude > 0 || consec_zeromv < 4))
return COPY_BLOCK;
if (is_skin && (motion_magnitude > 0 || consec_zeromv < 4)) return COPY_BLOCK;
// Avoid denoising for small block (unless motion is small).
// Small blocks are selected in variance partition (before encoding) and
// will typically lie on moving areas.
if (denoiser->denoising_level < kDenHigh &&
motion_magnitude > 16 && bs <= BLOCK_8X8)
if (denoiser->denoising_level < kDenHigh && motion_magnitude > 16 &&
bs <= BLOCK_8X8)
return COPY_BLOCK;
// If the best reference frame uses inter-prediction and there is enough of a
// difference in sum-squared-error, use it.
if (frame != INTRA_FRAME &&
ctx->newmv_sse != UINT_MAX &&
if (frame != INTRA_FRAME && ctx->newmv_sse != UINT_MAX &&
sse_diff > sse_diff_thresh(bs, increase_denoising, motion_magnitude)) {
mi->ref_frame[0] = ctx->best_reference_frame;
mi->mode = ctx->best_sse_inter_mode;
@@ -230,7 +222,7 @@ static VP9_DENOISER_DECISION perform_motion_compensation(VP9_DENOISER *denoiser,
ctx->newmv_sse = ctx->zeromv_sse;
// Bias to last reference.
if (frame != LAST_FRAME &&
((ctx->zeromv_lastref_sse < (5 * ctx->zeromv_sse) >> 2) ||
((ctx->zeromv_lastref_sse<(5 * ctx->zeromv_sse)>> 2) ||
denoiser->denoising_level >= kDenHigh)) {
frame = LAST_FRAME;
ctx->newmv_sse = ctx->zeromv_lastref_sse;
@@ -251,8 +243,7 @@ static VP9_DENOISER_DECISION perform_motion_compensation(VP9_DENOISER *denoiser,
*mi = saved_mi;
return COPY_BLOCK;
}
if (motion_magnitude >
(noise_motion_thresh(bs, increase_denoising) << 3)) {
if (motion_magnitude > (noise_motion_thresh(bs, increase_denoising) << 3)) {
// Restore everything to its original state
*mi = saved_mi;
return COPY_BLOCK;
@@ -268,37 +259,28 @@ static VP9_DENOISER_DECISION perform_motion_compensation(VP9_DENOISER *denoiser,
// struct.
filter_mbd->plane[0].pre[0].buf =
block_start(denoiser->running_avg_y[frame].y_buffer,
denoiser->running_avg_y[frame].y_stride,
mi_row, mi_col);
filter_mbd->plane[0].pre[0].stride =
denoiser->running_avg_y[frame].y_stride;
denoiser->running_avg_y[frame].y_stride, mi_row, mi_col);
filter_mbd->plane[0].pre[0].stride = denoiser->running_avg_y[frame].y_stride;
filter_mbd->plane[1].pre[0].buf =
block_start(denoiser->running_avg_y[frame].u_buffer,
denoiser->running_avg_y[frame].uv_stride,
mi_row, mi_col);
filter_mbd->plane[1].pre[0].stride =
denoiser->running_avg_y[frame].uv_stride;
block_start(denoiser->running_avg_y[frame].u_buffer,
denoiser->running_avg_y[frame].uv_stride, mi_row, mi_col);
filter_mbd->plane[1].pre[0].stride = denoiser->running_avg_y[frame].uv_stride;
filter_mbd->plane[2].pre[0].buf =
block_start(denoiser->running_avg_y[frame].v_buffer,
denoiser->running_avg_y[frame].uv_stride,
mi_row, mi_col);
filter_mbd->plane[2].pre[0].stride =
denoiser->running_avg_y[frame].uv_stride;
denoiser->running_avg_y[frame].uv_stride, mi_row, mi_col);
filter_mbd->plane[2].pre[0].stride = denoiser->running_avg_y[frame].uv_stride;
filter_mbd->plane[0].dst.buf =
block_start(denoiser->mc_running_avg_y.y_buffer,
denoiser->mc_running_avg_y.y_stride,
mi_row, mi_col);
denoiser->mc_running_avg_y.y_stride, mi_row, mi_col);
filter_mbd->plane[0].dst.stride = denoiser->mc_running_avg_y.y_stride;
filter_mbd->plane[1].dst.buf =
block_start(denoiser->mc_running_avg_y.u_buffer,
denoiser->mc_running_avg_y.uv_stride,
mi_row, mi_col);
denoiser->mc_running_avg_y.uv_stride, mi_row, mi_col);
filter_mbd->plane[1].dst.stride = denoiser->mc_running_avg_y.uv_stride;
filter_mbd->plane[2].dst.buf =
block_start(denoiser->mc_running_avg_y.v_buffer,
denoiser->mc_running_avg_y.uv_stride,
mi_row, mi_col);
denoiser->mc_running_avg_y.uv_stride, mi_row, mi_col);
filter_mbd->plane[2].dst.stride = denoiser->mc_running_avg_y.uv_stride;
vp9_build_inter_predictors_sby(filter_mbd, mi_row, mi_col, bs);
@@ -313,9 +295,8 @@ static VP9_DENOISER_DECISION perform_motion_compensation(VP9_DENOISER *denoiser,
return FILTER_BLOCK;
}
void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb,
int mi_row, int mi_col, BLOCK_SIZE bs,
PICK_MODE_CONTEXT *ctx,
void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb, int mi_row, int mi_col,
BLOCK_SIZE bs, PICK_MODE_CONTEXT *ctx,
VP9_DENOISER_DECISION *denoiser_decision) {
int mv_col, mv_row;
int motion_magnitude = 0;
@@ -325,8 +306,8 @@ void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb,
YV12_BUFFER_CONFIG avg = denoiser->running_avg_y[INTRA_FRAME];
YV12_BUFFER_CONFIG mc_avg = denoiser->mc_running_avg_y;
uint8_t *avg_start = block_start(avg.y_buffer, avg.y_stride, mi_row, mi_col);
uint8_t *mc_avg_start = block_start(mc_avg.y_buffer, mc_avg.y_stride,
mi_row, mi_col);
uint8_t *mc_avg_start =
block_start(mc_avg.y_buffer, mc_avg.y_stride, mi_row, mi_col);
struct buf_2d src = mb->plane[0].src;
int is_skin = 0;
int consec_zeromv = 0;
@@ -334,8 +315,7 @@ void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb,
mv_row = ctx->best_sse_mv.as_mv.row;
motion_magnitude = mv_row * mv_row + mv_col * mv_col;
if (cpi->use_skin_detection &&
bs <= BLOCK_32X32 &&
if (cpi->use_skin_detection && bs <= BLOCK_32X32 &&
denoiser->denoising_level < kDenHigh) {
int motion_level = (motion_magnitude < 16) ? 0 : 1;
// If motion for current block is small/zero, compute consec_zeromv for
@@ -343,7 +323,7 @@ void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb,
// consec_zeromv when current block has small/zero motion).
consec_zeromv = 0;
if (motion_level == 0) {
VP9_COMMON * const cm = &cpi->common;
VP9_COMMON *const cm = &cpi->common;
int j, i;
// Loop through the 8x8 sub-blocks.
const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64];
@@ -367,48 +347,35 @@ void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb,
}
}
// TODO(marpan): Compute skin detection over sub-blocks.
is_skin = vp9_compute_skin_block(mb->plane[0].src.buf,
mb->plane[1].src.buf,
mb->plane[2].src.buf,
mb->plane[0].src.stride,
mb->plane[1].src.stride,
bs,
consec_zeromv,
motion_level);
is_skin = vp9_compute_skin_block(
mb->plane[0].src.buf, mb->plane[1].src.buf, mb->plane[2].src.buf,
mb->plane[0].src.stride, mb->plane[1].src.stride, bs, consec_zeromv,
motion_level);
}
if (!is_skin &&
denoiser->denoising_level == kDenHigh) {
if (!is_skin && denoiser->denoising_level == kDenHigh) {
denoiser->increase_denoising = 1;
} else {
denoiser->increase_denoising = 0;
}
if (denoiser->denoising_level >= kDenLow)
decision = perform_motion_compensation(denoiser, mb, bs,
denoiser->increase_denoising,
mi_row, mi_col, ctx,
motion_magnitude,
is_skin,
&zeromv_filter,
consec_zeromv);
decision = perform_motion_compensation(
denoiser, mb, bs, denoiser->increase_denoising, mi_row, mi_col, ctx,
motion_magnitude, is_skin, &zeromv_filter, consec_zeromv);
if (decision == FILTER_BLOCK) {
decision = vp9_denoiser_filter(src.buf, src.stride,
mc_avg_start, mc_avg.y_stride,
avg_start, avg.y_stride,
denoiser->increase_denoising,
bs, motion_magnitude);
decision = vp9_denoiser_filter(
src.buf, src.stride, mc_avg_start, mc_avg.y_stride, avg_start,
avg.y_stride, denoiser->increase_denoising, bs, motion_magnitude);
}
if (decision == FILTER_BLOCK) {
vpx_convolve_copy(avg_start, avg.y_stride, src.buf, src.stride,
NULL, 0, NULL, 0,
num_4x4_blocks_wide_lookup[bs] << 2,
vpx_convolve_copy(avg_start, avg.y_stride, src.buf, src.stride, NULL, 0,
NULL, 0, num_4x4_blocks_wide_lookup[bs] << 2,
num_4x4_blocks_high_lookup[bs] << 2);
} else { // COPY_BLOCK
vpx_convolve_copy(src.buf, src.stride, avg_start, avg.y_stride,
NULL, 0, NULL, 0,
num_4x4_blocks_wide_lookup[bs] << 2,
vpx_convolve_copy(src.buf, src.stride, avg_start, avg.y_stride, NULL, 0,
NULL, 0, num_4x4_blocks_wide_lookup[bs] << 2,
num_4x4_blocks_high_lookup[bs] << 2);
}
*denoiser_decision = decision;
@@ -416,8 +383,8 @@ void vp9_denoiser_denoise(VP9_COMP *cpi, MACROBLOCK *mb,
*denoiser_decision = FILTER_ZEROMV_BLOCK;
}
static void copy_frame(YV12_BUFFER_CONFIG * const dest,
const YV12_BUFFER_CONFIG * const src) {
static void copy_frame(YV12_BUFFER_CONFIG *const dest,
const YV12_BUFFER_CONFIG *const src) {
int r;
const uint8_t *srcbuf = src->y_buffer;
uint8_t *destbuf = dest->y_buffer;
@@ -432,8 +399,8 @@ static void copy_frame(YV12_BUFFER_CONFIG * const dest,
}
}
static void swap_frame_buffer(YV12_BUFFER_CONFIG * const dest,
YV12_BUFFER_CONFIG * const src) {
static void swap_frame_buffer(YV12_BUFFER_CONFIG *const dest,
YV12_BUFFER_CONFIG *const src) {
uint8_t *tmp_buf = dest->y_buffer;
assert(dest->y_width == src->y_width);
assert(dest->y_height == src->y_height);
@@ -446,8 +413,7 @@ void vp9_denoiser_update_frame_info(VP9_DENOISER *denoiser,
FRAME_TYPE frame_type,
int refresh_alt_ref_frame,
int refresh_golden_frame,
int refresh_last_frame,
int resized) {
int refresh_last_frame, int resized) {
// Copy source into denoised reference buffers on KEY_FRAME or
// if the just encoded frame was resized.
if (frame_type == KEY_FRAME || resized != 0 || denoiser->reset) {
@@ -460,8 +426,7 @@ void vp9_denoiser_update_frame_info(VP9_DENOISER *denoiser,
}
// If more than one refresh occurs, must copy frame buffer.
if ((refresh_alt_ref_frame + refresh_golden_frame + refresh_last_frame)
> 1) {
if ((refresh_alt_ref_frame + refresh_golden_frame + refresh_last_frame) > 1) {
if (refresh_alt_ref_frame) {
copy_frame(&denoiser->running_avg_y[ALTREF_FRAME],
&denoiser->running_avg_y[INTRA_FRAME]);
@@ -503,8 +468,7 @@ void vp9_denoiser_update_frame_stats(MODE_INFO *mi, unsigned int sse,
if (mi->mv[0].as_int == 0 && sse < ctx->zeromv_sse) {
ctx->zeromv_sse = sse;
ctx->best_zeromv_reference_frame = mi->ref_frame[0];
if (mi->ref_frame[0] == LAST_FRAME)
ctx->zeromv_lastref_sse = sse;
if (mi->ref_frame[0] == LAST_FRAME) ctx->zeromv_lastref_sse = sse;
}
if (mi->mv[0].as_int != 0 && sse < ctx->newmv_sse) {
@@ -515,8 +479,8 @@ void vp9_denoiser_update_frame_stats(MODE_INFO *mi, unsigned int sse,
}
}
int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height,
int ssx, int ssy,
int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height, int ssx,
int ssy,
#if CONFIG_VP9_HIGHBITDEPTH
int use_highbitdepth,
#endif
@@ -541,8 +505,8 @@ int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height,
#endif
}
fail = vpx_alloc_frame_buffer(&denoiser->mc_running_avg_y, width, height,
ssx, ssy,
fail = vpx_alloc_frame_buffer(&denoiser->mc_running_avg_y, width, height, ssx,
ssy,
#if CONFIG_VP9_HIGHBITDEPTH
use_highbitdepth,
#endif
@@ -552,8 +516,7 @@ int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height,
return 1;
}
fail = vpx_alloc_frame_buffer(&denoiser->last_source, width, height,
ssx, ssy,
fail = vpx_alloc_frame_buffer(&denoiser->last_source, width, height, ssx, ssy,
#if CONFIG_VP9_HIGHBITDEPTH
use_highbitdepth,
#endif
@@ -586,8 +549,7 @@ void vp9_denoiser_free(VP9_DENOISER *denoiser) {
vpx_free_frame_buffer(&denoiser->last_source);
}
void vp9_denoiser_set_noise_level(VP9_DENOISER *denoiser,
int noise_level) {
void vp9_denoiser_set_noise_level(VP9_DENOISER *denoiser, int noise_level) {
denoiser->denoising_level = noise_level;
if (denoiser->denoising_level > kDenLowLow &&
denoiser->prev_denoising_level == kDenLowLow)

19
vp9/encoder/vp9_denoiser.h
View File

@@ -64,22 +64,20 @@ void vp9_denoiser_update_frame_info(VP9_DENOISER *denoiser,
FRAME_TYPE frame_type,
int refresh_alt_ref_frame,
int refresh_golden_frame,
int refresh_last_frame,
int resized);
int refresh_last_frame, int resized);
void vp9_denoiser_denoise(struct VP9_COMP *cpi, MACROBLOCK *mb,
int mi_row, int mi_col, BLOCK_SIZE bs,
PICK_MODE_CONTEXT *ctx ,
void vp9_denoiser_denoise(struct VP9_COMP *cpi, MACROBLOCK *mb, int mi_row,
int mi_col, BLOCK_SIZE bs, PICK_MODE_CONTEXT *ctx,
VP9_DENOISER_DECISION *denoiser_decision);
void vp9_denoiser_reset_frame_stats(PICK_MODE_CONTEXT *ctx);
void vp9_denoiser_update_frame_stats(MODE_INFO *mi,
unsigned int sse, PREDICTION_MODE mode,
void vp9_denoiser_update_frame_stats(MODE_INFO *mi, unsigned int sse,
PREDICTION_MODE mode,
PICK_MODE_CONTEXT *ctx);
int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height,
int ssx, int ssy,
int vp9_denoiser_alloc(VP9_DENOISER *denoiser, int width, int height, int ssx,
int ssy,
#if CONFIG_VP9_HIGHBITDEPTH
int use_highbitdepth,
#endif
@@ -97,8 +95,7 @@ static INLINE int total_adj_strong_thresh(BLOCK_SIZE bs,
void vp9_denoiser_free(VP9_DENOISER *denoiser);
void vp9_denoiser_set_noise_level(VP9_DENOISER *denoiser,
int noise_level);
void vp9_denoiser_set_noise_level(VP9_DENOISER *denoiser, int noise_level);
#ifdef __cplusplus
} // extern "C"

1381
vp9/encoder/vp9_encodeframe.c
View File

File diff suppressed because it is too large Load Diff

9
vp9/encoder/vp9_encodeframe.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_ENCODEFRAME_H_
#define VP9_ENCODER_VP9_ENCODEFRAME_H_
@@ -31,14 +30,14 @@ struct ThreadData;
#define VAR_HIST_SMALL_CUT_OFF 45
void vp9_setup_src_planes(struct macroblock *x,
const struct yv12_buffer_config *src,
int mi_row, int mi_col);
const struct yv12_buffer_config *src, int mi_row,
int mi_col);
void vp9_encode_frame(struct VP9_COMP *cpi);
void vp9_init_tile_data(struct VP9_COMP *cpi);
void vp9_encode_tile(struct VP9_COMP *cpi, struct ThreadData *td,
int tile_row, int tile_col);
void vp9_encode_tile(struct VP9_COMP *cpi, struct ThreadData *td, int tile_row,
int tile_col);
void vp9_set_variance_partition_thresholds(struct VP9_COMP *cpi, int q);

356
vp9/encoder/vp9_encodemb.c
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "./vp9_rtcd.h"
#include "./vpx_config.h"
#include "./vpx_dsp_rtcd.h"
@@ -51,29 +50,29 @@ void vp9_subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
}
typedef struct vp9_token_state {
int64_t error;
int rate;
int16_t next;
int16_t token;
tran_low_t qc;
tran_low_t dqc;
uint8_t best_index;
int64_t error;
int rate;
int16_t next;
int16_t token;
tran_low_t qc;
tran_low_t dqc;
uint8_t best_index;
} vp9_token_state;
static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] ={ {10, 6}, {8, 5}, };
static const int plane_rd_mult[REF_TYPES][PLANE_TYPES] = {
{ 10, 6 }, { 8, 5 },
};
#define UPDATE_RD_COST()\
{\
rd_cost0 = RDCOST(rdmult, rddiv, rate0, error0);\
rd_cost1 = RDCOST(rdmult, rddiv, rate1, error1);\
}
#define UPDATE_RD_COST() \
{ \
rd_cost0 = RDCOST(rdmult, rddiv, rate0, error0); \
rd_cost1 = RDCOST(rdmult, rddiv, rate1, error1); \
}
// This function is a place holder for now but may ultimately need
// to scan previous tokens to work out the correct context.
static int trellis_get_coeff_context(const int16_t *scan,
const int16_t *nb,
int idx, int token,
uint8_t *token_cache) {
static int trellis_get_coeff_context(const int16_t *scan, const int16_t *nb,
int idx, int token, uint8_t *token_cache) {
int bak = token_cache[scan[idx]], pt;
token_cache[scan[idx]] = vp9_pt_energy_class[token];
pt = get_coef_context(nb, token_cache, idx + 1);
@@ -81,8 +80,8 @@ static int trellis_get_coeff_context(const int16_t *scan,
return pt;
}
int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
TX_SIZE tx_size, int ctx) {
int vp9_optimize_b(MACROBLOCK *mb, int plane, int block, TX_SIZE tx_size,
int ctx) {
MACROBLOCKD *const xd = &mb->e_mbd;
struct macroblock_plane *const p = &mb->plane[plane];
struct macroblockd_plane *const pd = &xd->plane[plane];
@@ -96,8 +95,8 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
const PLANE_TYPE type = get_plane_type(plane);
const int default_eob = 16 << (tx_size << 1);
const int shift = (tx_size == TX_32X32);
const int16_t* const dequant_ptr = pd->dequant;
const uint8_t* const band_translate = get_band_translate(tx_size);
const int16_t *const dequant_ptr = pd->dequant;
const uint8_t *const band_translate = get_band_translate(tx_size);
const scan_order *const so = get_scan(xd, tx_size, type, block);
const int16_t *const scan = so->scan;
const int16_t *const nb = so->neighbors;
@@ -130,8 +129,7 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
tokens[eob][1] = tokens[eob][0];
for (i = 0; i < eob; i++)
token_cache[scan[i]] =
vp9_pt_energy_class[vp9_get_token(qcoeff[scan[i]])];
token_cache[scan[i]] = vp9_pt_energy_class[vp9_get_token(qcoeff[scan[i]])];
for (i = eob; i-- > 0;) {
int base_bits, d2, dx;
@@ -150,10 +148,12 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
if (next < default_eob) {
band = band_translate[i + 1];
pt = trellis_get_coeff_context(scan, nb, i, t0, token_cache);
rate0 += mb->token_costs[tx_size][type][ref][band][0][pt]
[tokens[next][0].token];
rate1 += mb->token_costs[tx_size][type][ref][band][0][pt]
[tokens[next][1].token];
rate0 +=
mb->token_costs[tx_size][type][ref][band][0][pt][tokens[next][0]
.token];
rate1 +=
mb->token_costs[tx_size][type][ref][band][0][pt][tokens[next][1]
.token];
}
UPDATE_RD_COST();
/* And pick the best. */
@@ -179,8 +179,8 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
rate1 = tokens[next][1].rate;
if ((abs(x) * dequant_ptr[rc != 0] > (abs(coeff[rc]) << shift)) &&
(abs(x) * dequant_ptr[rc != 0] < (abs(coeff[rc]) << shift) +
dequant_ptr[rc != 0]))
(abs(x) * dequant_ptr[rc != 0] <
(abs(coeff[rc]) << shift) + dequant_ptr[rc != 0]))
shortcut = 1;
else
shortcut = 0;
@@ -210,13 +210,15 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
band = band_translate[i + 1];
if (t0 != EOB_TOKEN) {
pt = trellis_get_coeff_context(scan, nb, i, t0, token_cache);
rate0 += mb->token_costs[tx_size][type][ref][band][!x][pt]
[tokens[next][0].token];
rate0 +=
mb->token_costs[tx_size][type][ref][band][!x][pt][tokens[next][0]
.token];
}
if (t1 != EOB_TOKEN) {
pt = trellis_get_coeff_context(scan, nb, i, t1, token_cache);
rate1 += mb->token_costs[tx_size][type][ref][band][!x][pt]
[tokens[next][1].token];
rate1 +=
mb->token_costs[tx_size][type][ref][band][!x][pt][tokens[next][1]
.token];
}
}
@@ -249,8 +251,7 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
// Account for the rounding difference in the dequantized coefficeint
// value when the quantization index is dropped from an even number
// to an odd number.
if (shift & x)
offset += (dequant_ptr[rc != 0] & 0x01);
if (shift & x) offset += (dequant_ptr[rc != 0] & 0x01);
if (sz == 0)
tokens[i][1].dqc = dqcoeff[rc] - offset;
@@ -316,9 +317,8 @@ int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
return final_eob;
}
static INLINE void fdct32x32(int rd_transform,
const int16_t *src, tran_low_t *dst,
int src_stride) {
static INLINE void fdct32x32(int rd_transform, const int16_t *src,
tran_low_t *dst, int src_stride) {
if (rd_transform)
vpx_fdct32x32_rd(src, dst, src_stride);
else
@@ -356,33 +356,31 @@ void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block, int row, int col,
highbd_fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vp9_highbd_quantize_fp_32x32(coeff, 1024, x->skip_block, p->zbin,
p->round_fp, p->quant_fp, p->quant_shift,
qcoeff, dqcoeff, pd->dequant,
eob, scan_order->scan,
scan_order->iscan);
qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
case TX_16X16:
vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
vp9_highbd_quantize_fp(coeff, 256, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
case TX_8X8:
vpx_highbd_fdct8x8(src_diff, coeff, diff_stride);
vp9_highbd_quantize_fp(coeff, 64, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
case TX_4X4:
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vp9_highbd_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
default:
assert(0);
default: assert(0);
}
return;
}
@@ -399,27 +397,22 @@ void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block, int row, int col,
case TX_16X16:
vpx_fdct16x16(src_diff, coeff, diff_stride);
vp9_quantize_fp(coeff, 256, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
p->quant_fp, p->quant_shift, qcoeff, dqcoeff, pd->dequant,
eob, scan_order->scan, scan_order->iscan);
break;
case TX_8X8:
vp9_fdct8x8_quant(src_diff, diff_stride, coeff, 64,
x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
vp9_fdct8x8_quant(src_diff, diff_stride, coeff, 64, x->skip_block,
p->zbin, p->round_fp, p->quant_fp, p->quant_shift,
qcoeff, dqcoeff, pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
case TX_4X4:
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vp9_quantize_fp(coeff, 16, x->skip_block, p->zbin, p->round_fp,
p->quant_fp, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
default:
assert(0);
p->quant_fp, p->quant_shift, qcoeff, dqcoeff, pd->dequant,
eob, scan_order->scan, scan_order->iscan);
break;
default: assert(0); break;
}
}
@@ -447,23 +440,22 @@ void vp9_xform_quant_dc(MACROBLOCK *x, int plane, int block, int row, int col,
case TX_16X16:
vpx_highbd_fdct16x16_1(src_diff, coeff, diff_stride);
vpx_highbd_quantize_dc(coeff, 256, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
p->quant_fp[0], qcoeff, dqcoeff, pd->dequant[0],
eob);
break;
case TX_8X8:
vpx_highbd_fdct8x8_1(src_diff, coeff, diff_stride);
vpx_highbd_quantize_dc(coeff, 64, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
p->quant_fp[0], qcoeff, dqcoeff, pd->dequant[0],
eob);
break;
case TX_4X4:
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vpx_highbd_quantize_dc(coeff, 16, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
p->quant_fp[0], qcoeff, dqcoeff, pd->dequant[0],
eob);
break;
default:
assert(0);
default: assert(0);
}
return;
}
@@ -472,31 +464,25 @@ void vp9_xform_quant_dc(MACROBLOCK *x, int plane, int block, int row, int col,
switch (tx_size) {
case TX_32X32:
vpx_fdct32x32_1(src_diff, coeff, diff_stride);
vpx_quantize_dc_32x32(coeff, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
vpx_quantize_dc_32x32(coeff, x->skip_block, p->round, p->quant_fp[0],
qcoeff, dqcoeff, pd->dequant[0], eob);
break;
case TX_16X16:
vpx_fdct16x16_1(src_diff, coeff, diff_stride);
vpx_quantize_dc(coeff, 256, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
vpx_quantize_dc(coeff, 256, x->skip_block, p->round, p->quant_fp[0],
qcoeff, dqcoeff, pd->dequant[0], eob);
break;
case TX_8X8:
vpx_fdct8x8_1(src_diff, coeff, diff_stride);
vpx_quantize_dc(coeff, 64, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
vpx_quantize_dc(coeff, 64, x->skip_block, p->round, p->quant_fp[0],
qcoeff, dqcoeff, pd->dequant[0], eob);
break;
case TX_4X4:
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vpx_quantize_dc(coeff, 16, x->skip_block, p->round,
p->quant_fp[0], qcoeff, dqcoeff,
pd->dequant[0], eob);
break;
default:
assert(0);
vpx_quantize_dc(coeff, 16, x->skip_block, p->round, p->quant_fp[0],
qcoeff, dqcoeff, pd->dequant[0], eob);
break;
default: assert(0); break;
}
}
@@ -516,37 +502,36 @@ void vp9_xform_quant(MACROBLOCK *x, int plane, int block, int row, int col,
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
switch (tx_size) {
switch (tx_size) {
case TX_32X32:
highbd_fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vpx_highbd_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin,
p->round, p->quant, p->quant_shift, qcoeff,
dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
dqcoeff, pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
case TX_16X16:
vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
vpx_highbd_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
case TX_8X8:
vpx_highbd_fdct8x8(src_diff, coeff, diff_stride);
vpx_highbd_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
case TX_4X4:
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vpx_highbd_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
break;
default:
assert(0);
default: assert(0);
}
return;
}
@@ -562,34 +547,28 @@ void vp9_xform_quant(MACROBLOCK *x, int plane, int block, int row, int col,
break;
case TX_16X16:
vpx_fdct16x16(src_diff, coeff, diff_stride);
vpx_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
vpx_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
case TX_8X8:
vpx_fdct8x8(src_diff, coeff, diff_stride);
vpx_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
vpx_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
case TX_4X4:
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
break;
default:
assert(0);
break;
default: assert(0); break;
}
}
static void encode_block(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg) {
struct encode_b_args *const args = arg;
MACROBLOCK *const x = args->x;
MACROBLOCKD *const xd = &x->e_mbd;
@@ -649,35 +628,32 @@ static void encode_block(int plane, int block, int row, int col,
*a = *l = p->eobs[block] > 0;
}
if (p->eobs[block])
*(args->skip) = 0;
if (p->eobs[block]) *(args->skip) = 0;
if (x->skip_encode || p->eobs[block] == 0)
return;
if (x->skip_encode || p->eobs[block] == 0) return;
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
switch (tx_size) {
case TX_32X32:
vp9_highbd_idct32x32_add(dqcoeff, dst, pd->dst.stride,
p->eobs[block], xd->bd);
vp9_highbd_idct32x32_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
xd->bd);
break;
case TX_16X16:
vp9_highbd_idct16x16_add(dqcoeff, dst, pd->dst.stride,
p->eobs[block], xd->bd);
vp9_highbd_idct16x16_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
xd->bd);
break;
case TX_8X8:
vp9_highbd_idct8x8_add(dqcoeff, dst, pd->dst.stride,
p->eobs[block], xd->bd);
vp9_highbd_idct8x8_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
xd->bd);
break;
case TX_4X4:
// this is like vp9_short_idct4x4 but has a special case around eob<=1
// which is significant (not just an optimization) for the lossless
// case.
x->highbd_itxm_add(dqcoeff, dst, pd->dst.stride,
p->eobs[block], xd->bd);
x->highbd_itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block],
xd->bd);
break;
default:
assert(0 && "Invalid transform size");
default: assert(0 && "Invalid transform size");
}
return;
}
@@ -699,15 +675,13 @@ static void encode_block(int plane, int block, int row, int col,
// case.
x->itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
break;
default:
assert(0 && "Invalid transform size");
break;
default: assert(0 && "Invalid transform size"); break;
}
}
static void encode_block_pass1(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *arg) {
MACROBLOCK *const x = (MACROBLOCK *)arg;
MACROBLOCKD *const xd = &x->e_mbd;
struct macroblock_plane *const p = &x->plane[plane];
@@ -721,8 +695,8 @@ static void encode_block_pass1(int plane, int block, int row, int col,
if (p->eobs[block] > 0) {
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
x->highbd_itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block], xd->bd);
return;
x->highbd_itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block], xd->bd);
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
x->itxm_add(dqcoeff, dst, pd->dst.stride, p->eobs[block]);
@@ -739,23 +713,21 @@ void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
MACROBLOCKD *const xd = &x->e_mbd;
struct optimize_ctx ctx;
MODE_INFO *mi = xd->mi[0];
struct encode_b_args arg = {x, 1, NULL, NULL, &mi->skip};
struct encode_b_args arg = { x, 1, NULL, NULL, &mi->skip };
int plane;
mi->skip = 1;
if (x->skip)
return;
if (x->skip) return;
for (plane = 0; plane < MAX_MB_PLANE; ++plane) {
if (!x->skip_recode)
vp9_subtract_plane(x, bsize, plane);
if (!x->skip_recode) vp9_subtract_plane(x, bsize, plane);
if (x->optimize && (!x->skip_recode || !x->skip_optimize)) {
const struct macroblockd_plane* const pd = &xd->plane[plane];
const struct macroblockd_plane *const pd = &xd->plane[plane];
const TX_SIZE tx_size = plane ? get_uv_tx_size(mi, pd) : mi->tx_size;
vp9_get_entropy_contexts(bsize, tx_size, pd,
ctx.ta[plane], ctx.tl[plane]);
vp9_get_entropy_contexts(bsize, tx_size, pd, ctx.ta[plane],
ctx.tl[plane]);
arg.enable_coeff_opt = 1;
} else {
arg.enable_coeff_opt = 0;
@@ -769,9 +741,9 @@ void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize) {
}
void vp9_encode_block_intra(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct encode_b_args* const args = arg;
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *arg) {
struct encode_b_args *const args = arg;
MACROBLOCK *const x = args->x;
MACROBLOCKD *const xd = &x->e_mbd;
MODE_INFO *mi = xd->mi[0];
@@ -817,16 +789,16 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
}
vp9_predict_intra_block(xd, bwl, tx_size, mode, x->skip_encode ? src : dst,
x->skip_encode ? src_stride : dst_stride,
dst, dst_stride, col, row, plane);
x->skip_encode ? src_stride : dst_stride, dst,
dst_stride, col, row, plane);
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
switch (tx_size) {
case TX_32X32:
if (!x->skip_recode) {
vpx_highbd_subtract_block(32, 32, src_diff, diff_stride,
src, src_stride, dst, dst_stride, xd->bd);
vpx_highbd_subtract_block(32, 32, src_diff, diff_stride, src,
src_stride, dst, dst_stride, xd->bd);
highbd_fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vpx_highbd_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin,
p->round, p->quant, p->quant_shift,
@@ -839,34 +811,34 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
break;
case TX_16X16:
if (!x->skip_recode) {
vpx_highbd_subtract_block(16, 16, src_diff, diff_stride,
src, src_stride, dst, dst_stride, xd->bd);
vpx_highbd_subtract_block(16, 16, src_diff, diff_stride, src,
src_stride, dst, dst_stride, xd->bd);
if (tx_type == DCT_DCT)
vpx_highbd_fdct16x16(src_diff, coeff, diff_stride);
else
vp9_highbd_fht16x16(src_diff, coeff, diff_stride, tx_type);
vpx_highbd_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob) {
vp9_highbd_iht16x16_add(tx_type, dqcoeff, dst, dst_stride,
*eob, xd->bd);
vp9_highbd_iht16x16_add(tx_type, dqcoeff, dst, dst_stride, *eob,
xd->bd);
}
break;
case TX_8X8:
if (!x->skip_recode) {
vpx_highbd_subtract_block(8, 8, src_diff, diff_stride,
src, src_stride, dst, dst_stride, xd->bd);
vpx_highbd_subtract_block(8, 8, src_diff, diff_stride, src,
src_stride, dst, dst_stride, xd->bd);
if (tx_type == DCT_DCT)
vpx_highbd_fdct8x8(src_diff, coeff, diff_stride);
else
vp9_highbd_fht8x8(src_diff, coeff, diff_stride, tx_type);
vpx_highbd_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob) {
vp9_highbd_iht8x8_add(tx_type, dqcoeff, dst, dst_stride, *eob,
@@ -875,16 +847,16 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
break;
case TX_4X4:
if (!x->skip_recode) {
vpx_highbd_subtract_block(4, 4, src_diff, diff_stride,
src, src_stride, dst, dst_stride, xd->bd);
vpx_highbd_subtract_block(4, 4, src_diff, diff_stride, src,
src_stride, dst, dst_stride, xd->bd);
if (tx_type != DCT_DCT)
vp9_highbd_fht4x4(src_diff, coeff, diff_stride, tx_type);
else
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vpx_highbd_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob,
scan_order->scan, scan_order->iscan);
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
}
if (!x->skip_encode && *eob) {
@@ -898,12 +870,9 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
}
}
break;
default:
assert(0);
return;
default: assert(0); return;
}
if (*eob)
*(args->skip) = 0;
if (*eob) *(args->skip) = 0;
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
@@ -911,8 +880,8 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
switch (tx_size) {
case TX_32X32:
if (!x->skip_recode) {
vpx_subtract_block(32, 32, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
vpx_subtract_block(32, 32, src_diff, diff_stride, src, src_stride, dst,
dst_stride);
fdct32x32(x->use_lp32x32fdct, src_diff, coeff, diff_stride);
vpx_quantize_b_32x32(coeff, 1024, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
@@ -920,20 +889,19 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
scan_order->iscan);
}
if (args->enable_coeff_opt && !x->skip_recode) {
*a = *l = vp9_optimize_b(x, plane, block, tx_size, entropy_ctx) > 0;
*a = *l = vp9_optimize_b(x, plane, block, tx_size, entropy_ctx) > 0;
}
if (!x->skip_encode && *eob)
vp9_idct32x32_add(dqcoeff, dst, dst_stride, *eob);
break;
case TX_16X16:
if (!x->skip_recode) {
vpx_subtract_block(16, 16, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
vpx_subtract_block(16, 16, src_diff, diff_stride, src, src_stride, dst,
dst_stride);
vp9_fht16x16(src_diff, coeff, diff_stride, tx_type);
vpx_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round,
p->quant, p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
vpx_quantize_b(coeff, 256, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
}
if (args->enable_coeff_opt && !x->skip_recode) {
*a = *l = vp9_optimize_b(x, plane, block, tx_size, entropy_ctx) > 0;
@@ -943,13 +911,12 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
break;
case TX_8X8:
if (!x->skip_recode) {
vpx_subtract_block(8, 8, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
vpx_subtract_block(8, 8, src_diff, diff_stride, src, src_stride, dst,
dst_stride);
vp9_fht8x8(src_diff, coeff, diff_stride, tx_type);
vpx_quantize_b(coeff, 64, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
p->quant_shift, qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
}
if (args->enable_coeff_opt && !x->skip_recode) {
*a = *l = vp9_optimize_b(x, plane, block, tx_size, entropy_ctx) > 0;
@@ -959,16 +926,15 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
break;
case TX_4X4:
if (!x->skip_recode) {
vpx_subtract_block(4, 4, src_diff, diff_stride,
src, src_stride, dst, dst_stride);
vpx_subtract_block(4, 4, src_diff, diff_stride, src, src_stride, dst,
dst_stride);
if (tx_type != DCT_DCT)
vp9_fht4x4(src_diff, coeff, diff_stride, tx_type);
else
x->fwd_txm4x4(src_diff, coeff, diff_stride);
vpx_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant,
p->quant_shift, qcoeff, dqcoeff,
pd->dequant, eob, scan_order->scan,
scan_order->iscan);
p->quant_shift, qcoeff, dqcoeff, pd->dequant, eob,
scan_order->scan, scan_order->iscan);
}
if (args->enable_coeff_opt && !x->skip_recode) {
*a = *l = vp9_optimize_b(x, plane, block, tx_size, entropy_ctx) > 0;
@@ -983,27 +949,23 @@ void vp9_encode_block_intra(int plane, int block, int row, int col,
vp9_iht4x4_16_add(dqcoeff, dst, dst_stride, tx_type);
}
break;
default:
assert(0);
break;
default: assert(0); break;
}
if (*eob)
*(args->skip) = 0;
if (*eob) *(args->skip) = 0;
}
void vp9_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane,
int enable_optimize_b) {
const MACROBLOCKD *const xd = &x->e_mbd;
struct optimize_ctx ctx;
struct encode_b_args arg = {x, enable_optimize_b,
ctx.ta[plane], ctx.tl[plane],
&xd->mi[0]->skip};
struct encode_b_args arg = { x, enable_optimize_b, ctx.ta[plane],
ctx.tl[plane], &xd->mi[0]->skip };
if (enable_optimize_b && x->optimize &&
(!x->skip_recode || !x->skip_optimize)) {
const struct macroblockd_plane* const pd = &xd->plane[plane];
const TX_SIZE tx_size = plane ? get_uv_tx_size(xd->mi[0], pd) :
xd->mi[0]->tx_size;
const struct macroblockd_plane *const pd = &xd->plane[plane];
const TX_SIZE tx_size =
plane ? get_uv_tx_size(xd->mi[0], pd) : xd->mi[0]->tx_size;
vp9_get_entropy_contexts(bsize, tx_size, pd, ctx.ta[plane], ctx.tl[plane]);
} else {
arg.enable_coeff_opt = 0;

7
vp9/encoder/vp9_encodemb.h
View File

@@ -25,8 +25,8 @@ struct encode_b_args {
ENTROPY_CONTEXT *tl;
int8_t *skip;
};
int vp9_optimize_b(MACROBLOCK *mb, int plane, int block,
TX_SIZE tx_size, int ctx);
int vp9_optimize_b(MACROBLOCK *mb, int plane, int block, TX_SIZE tx_size,
int ctx);
void vp9_encode_sb(MACROBLOCK *x, BLOCK_SIZE bsize);
void vp9_encode_sby_pass1(MACROBLOCK *x, BLOCK_SIZE bsize);
void vp9_xform_quant_fp(MACROBLOCK *x, int plane, int block, int row, int col,
@@ -39,8 +39,7 @@ void vp9_xform_quant(MACROBLOCK *x, int plane, int block, int row, int col,
void vp9_subtract_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane);
void vp9_encode_block_intra(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg);
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg);
void vp9_encode_intra_block_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane,
int enable_optimize_b);

57
vp9/encoder/vp9_encodemv.c
View File

@@ -30,15 +30,15 @@ void vp9_entropy_mv_init(void) {
vp9_tokens_from_tree(mv_fp_encodings, vp9_mv_fp_tree);
}
static void encode_mv_component(vpx_writer* w, int comp,
const nmv_component* mvcomp, int usehp) {
static void encode_mv_component(vpx_writer *w, int comp,
const nmv_component *mvcomp, int usehp) {
int offset;
const int sign = comp < 0;
const int mag = sign ? -comp : comp;
const int mv_class = vp9_get_mv_class(mag - 1, &offset);
const int d = offset >> 3; // int mv data
const int fr = (offset >> 1) & 3; // fractional mv data
const int hp = offset & 1; // high precision mv data
const int d = offset >> 3; // int mv data
const int fr = (offset >> 1) & 3; // fractional mv data
const int hp = offset & 1; // high precision mv data
assert(comp != 0);
@@ -56,24 +56,21 @@ static void encode_mv_component(vpx_writer* w, int comp,
} else {
int i;
const int n = mv_class + CLASS0_BITS - 1; // number of bits
for (i = 0; i < n; ++i)
vpx_write(w, (d >> i) & 1, mvcomp->bits[i]);
for (i = 0; i < n; ++i) vpx_write(w, (d >> i) & 1, mvcomp->bits[i]);
}
// Fractional bits
vp9_write_token(w, vp9_mv_fp_tree,
mv_class == MV_CLASS_0 ? mvcomp->class0_fp[d] : mvcomp->fp,
mv_class == MV_CLASS_0 ? mvcomp->class0_fp[d] : mvcomp->fp,
&mv_fp_encodings[fr]);
// High precision bit
if (usehp)
vpx_write(w, hp,
mv_class == MV_CLASS_0 ? mvcomp->class0_hp : mvcomp->hp);
vpx_write(w, hp, mv_class == MV_CLASS_0 ? mvcomp->class0_hp : mvcomp->hp);
}
static void build_nmv_component_cost_table(int *mvcost,
const nmv_component* const mvcomp,
const nmv_component *const mvcomp,
int usehp) {
int sign_cost[2], class_cost[MV_CLASSES], class0_cost[CLASS0_SIZE];
int bits_cost[MV_OFFSET_BITS][2];
@@ -107,12 +104,12 @@ static void build_nmv_component_cost_table(int *mvcost,
int d, e, f;
int cost = class_cost[MV_CLASS_0];
int v = o + 1;
d = (o >> 3); /* int mv data */
f = (o >> 1) & 3; /* fractional pel mv data */
d = (o >> 3); /* int mv data */
f = (o >> 1) & 3; /* fractional pel mv data */
cost += class0_cost[d];
cost += class0_fp_cost[d][f];
if (usehp) {
e = (o & 1); /* high precision mv data */
e = (o & 1); /* high precision mv data */
cost += class0_hp_cost[e];
}
mvcost[v] = cost + sign_cost[0];
@@ -123,9 +120,8 @@ static void build_nmv_component_cost_table(int *mvcost,
for (d = 0; d < (1 << c); ++d) {
int f;
int whole_cost = class_cost[c];
int b = c + CLASS0_BITS - 1; /* number of bits */
for (i = 0; i < b; ++i)
whole_cost += bits_cost[i][((d >> i) & 1)];
int b = c + CLASS0_BITS - 1; /* number of bits */
for (i = 0; i < b; ++i) whole_cost += bits_cost[i][((d >> i) & 1)];
for (f = 0; f < 4; ++f) {
int cost = whole_cost + fp_cost[f];
int v = (CLASS0_SIZE << (c + 2)) + d * 8 + f * 2 /* + e */ + 1;
@@ -163,8 +159,8 @@ static int update_mv(vpx_writer *w, const unsigned int ct[2], vpx_prob *cur_p,
static void write_mv_update(const vpx_tree_index *tree,
vpx_prob probs[/*n - 1*/],
const unsigned int counts[/*n - 1*/],
int n, vpx_writer *w) {
const unsigned int counts[/*n - 1*/], int n,
vpx_writer *w) {
int i;
unsigned int branch_ct[32][2];
@@ -214,11 +210,9 @@ void vp9_write_nmv_probs(VP9_COMMON *cm, int usehp, vpx_writer *w,
}
}
void vp9_encode_mv(VP9_COMP* cpi, vpx_writer* w,
const MV* mv, const MV* ref,
const nmv_context* mvctx, int usehp) {
const MV diff = {mv->row - ref->row,
mv->col - ref->col};
void vp9_encode_mv(VP9_COMP *cpi, vpx_writer *w, const MV *mv, const MV *ref,
const nmv_context *mvctx, int usehp) {
const MV diff = { mv->row - ref->row, mv->col - ref->col };
const MV_JOINT_TYPE j = vp9_get_mv_joint(&diff);
usehp = usehp && use_mv_hp(ref);
@@ -238,21 +232,20 @@ void vp9_encode_mv(VP9_COMP* cpi, vpx_writer* w,
}
void vp9_build_nmv_cost_table(int *mvjoint, int *mvcost[2],
const nmv_context* ctx, int usehp) {
const nmv_context *ctx, int usehp) {
vp9_cost_tokens(mvjoint, ctx->joints, vp9_mv_joint_tree);
build_nmv_component_cost_table(mvcost[0], &ctx->comps[0], usehp);
build_nmv_component_cost_table(mvcost[1], &ctx->comps[1], usehp);
}
static void inc_mvs(const MODE_INFO *mi, const MB_MODE_INFO_EXT *mbmi_ext,
const int_mv mvs[2],
nmv_context_counts *counts) {
const int_mv mvs[2], nmv_context_counts *counts) {
int i;
for (i = 0; i < 1 + has_second_ref(mi); ++i) {
const MV *ref = &mbmi_ext->ref_mvs[mi->ref_frame[i]][0].as_mv;
const MV diff = {mvs[i].as_mv.row - ref->row,
mvs[i].as_mv.col - ref->col};
const MV diff = { mvs[i].as_mv.row - ref->row,
mvs[i].as_mv.col - ref->col };
vp9_inc_mv(&diff, counts);
}
}
@@ -275,8 +268,6 @@ void vp9_update_mv_count(ThreadData *td) {
}
}
} else {
if (mi->mode == NEWMV)
inc_mvs(mi, mbmi_ext, mi->mv, &td->counts->mv);
if (mi->mode == NEWMV) inc_mvs(mi, mbmi_ext, mi->mv, &td->counts->mv);
}
}

7
vp9/encoder/vp9_encodemv.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_ENCODEMV_H_
#define VP9_ENCODER_VP9_ENCODEMV_H_
@@ -23,11 +22,11 @@ void vp9_entropy_mv_init(void);
void vp9_write_nmv_probs(VP9_COMMON *cm, int usehp, vpx_writer *w,
nmv_context_counts *const counts);
void vp9_encode_mv(VP9_COMP *cpi, vpx_writer* w, const MV* mv, const MV* ref,
const nmv_context* mvctx, int usehp);
void vp9_encode_mv(VP9_COMP *cpi, vpx_writer *w, const MV *mv, const MV *ref,
const nmv_context *mvctx, int usehp);
void vp9_build_nmv_cost_table(int *mvjoint, int *mvcost[2],
const nmv_context* mvctx, int usehp);
const nmv_context *mvctx, int usehp);
void vp9_update_mv_count(ThreadData *td);

1739
vp9/encoder/vp9_encoder.c
View File

File diff suppressed because it is too large Load Diff

73
vp9/encoder/vp9_encoder.h
View File

@@ -72,7 +72,6 @@ typedef struct {
FRAME_CONTEXT fc;
} CODING_CONTEXT;
typedef enum {
// encode_breakout is disabled.
ENCODE_BREAKOUT_DISABLED = 0,
@@ -83,10 +82,10 @@ typedef enum {
} ENCODE_BREAKOUT_TYPE;
typedef enum {
NORMAL = 0,
FOURFIVE = 1,
THREEFIVE = 2,
ONETWO = 3
NORMAL = 0,
FOURFIVE = 1,
THREEFIVE = 2,
ONETWO = 3
} VPX_SCALING;
typedef enum {
@@ -106,7 +105,7 @@ typedef enum {
} MODE;
typedef enum {
FRAMEFLAGS_KEY = 1 << 0,
FRAMEFLAGS_KEY = 1 << 0,
FRAMEFLAGS_GOLDEN = 1 << 1,
FRAMEFLAGS_ALTREF = 1 << 2,
} FRAMETYPE_FLAGS;
@@ -129,14 +128,14 @@ typedef enum {
typedef struct VP9EncoderConfig {
BITSTREAM_PROFILE profile;
vpx_bit_depth_t bit_depth; // Codec bit-depth.
int width; // width of data passed to the compressor
int height; // height of data passed to the compressor
int width; // width of data passed to the compressor
int height; // height of data passed to the compressor
unsigned int input_bit_depth; // Input bit depth.
double init_framerate; // set to passed in framerate
int64_t target_bandwidth; // bandwidth to be used in bits per second
double init_framerate; // set to passed in framerate
int64_t target_bandwidth; // bandwidth to be used in bits per second
int noise_sensitivity; // pre processing blur: recommendation 0
int sharpness; // sharpening output: recommendation 0:
int sharpness; // sharpening output: recommendation 0:
int speed;
// maximum allowed bitrate for any intra frame in % of bitrate target.
unsigned int rc_max_intra_bitrate_pct;
@@ -188,7 +187,7 @@ typedef struct VP9EncoderConfig {
int frame_periodic_boost;
// two pass datarate control
int two_pass_vbrbias; // two pass datarate control tweaks
int two_pass_vbrbias; // two pass datarate control tweaks
int two_pass_vbrmin_section;
int two_pass_vbrmax_section;
// END DATARATE CONTROL OPTIONS
@@ -289,15 +288,10 @@ typedef struct ActiveMap {
unsigned char *map;
} ActiveMap;
typedef enum {
Y,
U,
V,
ALL
} STAT_TYPE;
typedef enum { Y, U, V, ALL } STAT_TYPE;
typedef struct IMAGE_STAT {
double stat[ALL+1];
double stat[ALL + 1];
double worst;
} ImageStat;
@@ -333,7 +327,7 @@ typedef struct {
uint64_t max_luma_sample_rate;
uint32_t max_luma_picture_size;
double average_bitrate; // in kilobits per second
double max_cpb_size; // in kilobits
double max_cpb_size; // in kilobits
double compression_ratio;
uint8_t max_col_tiles;
uint32_t min_altref_distance;
@@ -375,8 +369,8 @@ typedef struct VP9_COMP {
DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]);
VP9_COMMON common;
VP9EncoderConfig oxcf;
struct lookahead_ctx *lookahead;
struct lookahead_entry *alt_ref_source;
struct lookahead_ctx *lookahead;
struct lookahead_entry *alt_ref_source;
YV12_BUFFER_CONFIG *Source;
YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
@@ -439,11 +433,11 @@ typedef struct VP9_COMP {
int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE];
struct vpx_codec_pkt_list *output_pkt_list;
struct vpx_codec_pkt_list *output_pkt_list;
MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
int mbgraph_n_frames; // number of frames filled in the above
int static_mb_pct; // % forced skip mbs by segmentation
int mbgraph_n_frames; // number of frames filled in the above
int static_mb_pct; // % forced skip mbs by segmentation
int ref_frame_flags;
SPEED_FEATURES sf;
@@ -500,7 +494,7 @@ typedef struct VP9_COMP {
double total_blockiness;
double worst_blockiness;
int bytes;
int bytes;
double summed_quality;
double summed_weights;
double summedp_quality;
@@ -606,15 +600,15 @@ void vp9_remove_compressor(VP9_COMP *cpi);
void vp9_change_config(VP9_COMP *cpi, const VP9EncoderConfig *oxcf);
// receive a frames worth of data. caller can assume that a copy of this
// frame is made and not just a copy of the pointer..
// receive a frames worth of data. caller can assume that a copy of this
// frame is made and not just a copy of the pointer..
int vp9_receive_raw_frame(VP9_COMP *cpi, unsigned int frame_flags,
YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
int64_t end_time_stamp);
int vp9_get_compressed_data(VP9_COMP *cpi, unsigned int *frame_flags,
size_t *size, uint8_t *dest,
int64_t *time_stamp, int64_t *time_end, int flush);
size_t *size, uint8_t *dest, int64_t *time_stamp,
int64_t *time_end, int flush);
int vp9_get_preview_raw_frame(VP9_COMP *cpi, YV12_BUFFER_CONFIG *dest,
vp9_ppflags_t *flags);
@@ -635,8 +629,8 @@ int vp9_set_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
int vp9_get_active_map(VP9_COMP *cpi, unsigned char *map, int rows, int cols);
int vp9_set_internal_size(VP9_COMP *cpi,
VPX_SCALING horiz_mode, VPX_SCALING vert_mode);
int vp9_set_internal_size(VP9_COMP *cpi, VPX_SCALING horiz_mode,
VPX_SCALING vert_mode);
int vp9_set_size_literal(VP9_COMP *cpi, unsigned int width,
unsigned int height);
@@ -646,8 +640,7 @@ void vp9_set_svc(VP9_COMP *cpi, int use_svc);
int vp9_get_quantizer(struct VP9_COMP *cpi);
static INLINE int frame_is_kf_gf_arf(const VP9_COMP *cpi) {
return frame_is_intra_only(&cpi->common) ||
cpi->refresh_alt_ref_frame ||
return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
(cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
}
@@ -673,8 +666,8 @@ static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
VP9_COMMON *const cm = &cpi->common;
const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
return
buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf : NULL;
return buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf
: NULL;
}
static INLINE int get_token_alloc(int mb_rows, int mb_cols) {
@@ -737,10 +730,10 @@ static INLINE int is_altref_enabled(const VP9_COMP *const cpi) {
static INLINE void set_ref_ptrs(VP9_COMMON *cm, MACROBLOCKD *xd,
MV_REFERENCE_FRAME ref0,
MV_REFERENCE_FRAME ref1) {
xd->block_refs[0] = &cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME
: 0];
xd->block_refs[1] = &cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME
: 0];
xd->block_refs[0] =
&cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME : 0];
xd->block_refs[1] =
&cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME : 0];
}
static INLINE int get_chessboard_index(const int frame_index) {

17
vp9/encoder/vp9_ethread.c
View File

@@ -43,10 +43,10 @@ static int enc_worker_hook(EncWorkerData *const thread_data, void *unused) {
const int tile_rows = 1 << cm->log2_tile_rows;
int t;
(void) unused;
(void)unused;
for (t = thread_data->start; t < tile_rows * tile_cols;
t += cpi->num_workers) {
t += cpi->num_workers) {
int tile_row = t / tile_cols;
int tile_col = t % tile_cols;
@@ -63,8 +63,8 @@ static int get_max_tile_cols(VP9_COMP *cpi) {
int log2_tile_cols;
vp9_get_tile_n_bits(mi_cols, &min_log2_tile_cols, &max_log2_tile_cols);
log2_tile_cols = clamp(cpi->oxcf.tile_columns,
min_log2_tile_cols, max_log2_tile_cols);
log2_tile_cols =
clamp(cpi->oxcf.tile_columns, min_log2_tile_cols, max_log2_tile_cols);
return (1 << log2_tile_cols);
}
@@ -92,8 +92,7 @@ void vp9_encode_tiles_mt(VP9_COMP *cpi) {
vpx_malloc(allocated_workers * sizeof(*cpi->workers)));
CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
vpx_calloc(allocated_workers,
sizeof(*cpi->tile_thr_data)));
vpx_calloc(allocated_workers, sizeof(*cpi->tile_thr_data)));
for (i = 0; i < allocated_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
@@ -140,7 +139,7 @@ void vp9_encode_tiles_mt(VP9_COMP *cpi) {
worker->hook = (VPxWorkerHook)enc_worker_hook;
worker->data1 = &cpi->tile_thr_data[i];
worker->data2 = NULL;
thread_data = (EncWorkerData*)worker->data1;
thread_data = (EncWorkerData *)worker->data1;
// Before encoding a frame, copy the thread data from cpi.
if (thread_data->td != &cpi->td) {
@@ -173,7 +172,7 @@ void vp9_encode_tiles_mt(VP9_COMP *cpi) {
// Encode a frame
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData*)worker->data1;
EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
// Set the starting tile for each thread.
thread_data->start = i;
@@ -192,7 +191,7 @@ void vp9_encode_tiles_mt(VP9_COMP *cpi) {
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData*)worker->data1;
EncWorkerData *const thread_data = (EncWorkerData *)worker->data1;
// Accumulate counters.
if (i < cpi->num_workers - 1) {

76
vp9/encoder/vp9_extend.c
View File

@@ -16,8 +16,7 @@
#include "vp9/encoder/vp9_extend.h"
static void copy_and_extend_plane(const uint8_t *src, int src_pitch,
uint8_t *dst, int dst_pitch,
int w, int h,
uint8_t *dst, int dst_pitch, int w, int h,
int extend_top, int extend_left,
int extend_bottom, int extend_right) {
int i, linesize;
@@ -43,7 +42,7 @@ static void copy_and_extend_plane(const uint8_t *src, int src_pitch,
src_ptr1 = dst - extend_left;
src_ptr2 = dst + dst_pitch * (h - 1) - extend_left;
dst_ptr1 = dst + dst_pitch * (-extend_top) - extend_left;
dst_ptr2 = dst + dst_pitch * (h) - extend_left;
dst_ptr2 = dst + dst_pitch * (h)-extend_left;
linesize = extend_left + extend_right + w;
for (i = 0; i < extend_top; i++) {
@@ -59,9 +58,8 @@ static void copy_and_extend_plane(const uint8_t *src, int src_pitch,
#if CONFIG_VP9_HIGHBITDEPTH
static void highbd_copy_and_extend_plane(const uint8_t *src8, int src_pitch,
uint8_t *dst8, int dst_pitch,
int w, int h,
int extend_top, int extend_left,
uint8_t *dst8, int dst_pitch, int w,
int h, int extend_top, int extend_left,
int extend_bottom, int extend_right) {
int i, linesize;
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
@@ -88,7 +86,7 @@ static void highbd_copy_and_extend_plane(const uint8_t *src8, int src_pitch,
src_ptr1 = dst - extend_left;
src_ptr2 = dst + dst_pitch * (h - 1) - extend_left;
dst_ptr1 = dst + dst_pitch * (-extend_top) - extend_left;
dst_ptr2 = dst + dst_pitch * (h) - extend_left;
dst_ptr2 = dst + dst_pitch * (h)-extend_left;
linesize = extend_left + extend_right + w;
for (i = 0; i < extend_top; i++) {
@@ -127,51 +125,46 @@ void vp9_copy_and_extend_frame(const YV12_BUFFER_CONFIG *src,
#if CONFIG_VP9_HIGHBITDEPTH
if (src->flags & YV12_FLAG_HIGHBITDEPTH) {
highbd_copy_and_extend_plane(src->y_buffer, src->y_stride,
dst->y_buffer, dst->y_stride,
src->y_crop_width, src->y_crop_height,
et_y, el_y, eb_y, er_y);
highbd_copy_and_extend_plane(src->y_buffer, src->y_stride, dst->y_buffer,
dst->y_stride, src->y_crop_width,
src->y_crop_height, et_y, el_y, eb_y, er_y);
highbd_copy_and_extend_plane(src->u_buffer, src->uv_stride,
dst->u_buffer, dst->uv_stride,
src->uv_crop_width, src->uv_crop_height,
et_uv, el_uv, eb_uv, er_uv);
highbd_copy_and_extend_plane(
src->u_buffer, src->uv_stride, dst->u_buffer, dst->uv_stride,
src->uv_crop_width, src->uv_crop_height, et_uv, el_uv, eb_uv, er_uv);
highbd_copy_and_extend_plane(src->v_buffer, src->uv_stride,
dst->v_buffer, dst->uv_stride,
src->uv_crop_width, src->uv_crop_height,
et_uv, el_uv, eb_uv, er_uv);
highbd_copy_and_extend_plane(
src->v_buffer, src->uv_stride, dst->v_buffer, dst->uv_stride,
src->uv_crop_width, src->uv_crop_height, et_uv, el_uv, eb_uv, er_uv);
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
copy_and_extend_plane(src->y_buffer, src->y_stride,
dst->y_buffer, dst->y_stride,
src->y_crop_width, src->y_crop_height,
copy_and_extend_plane(src->y_buffer, src->y_stride, dst->y_buffer,
dst->y_stride, src->y_crop_width, src->y_crop_height,
et_y, el_y, eb_y, er_y);
copy_and_extend_plane(src->u_buffer, src->uv_stride,
dst->u_buffer, dst->uv_stride,
src->uv_crop_width, src->uv_crop_height,
copy_and_extend_plane(src->u_buffer, src->uv_stride, dst->u_buffer,
dst->uv_stride, src->uv_crop_width, src->uv_crop_height,
et_uv, el_uv, eb_uv, er_uv);
copy_and_extend_plane(src->v_buffer, src->uv_stride,
dst->v_buffer, dst->uv_stride,
src->uv_crop_width, src->uv_crop_height,
copy_and_extend_plane(src->v_buffer, src->uv_stride, dst->v_buffer,
dst->uv_stride, src->uv_crop_width, src->uv_crop_height,
et_uv, el_uv, eb_uv, er_uv);
}
void vp9_copy_and_extend_frame_with_rect(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst,
int srcy, int srcx,
int srch, int srcw) {
YV12_BUFFER_CONFIG *dst, int srcy,
int srcx, int srch, int srcw) {
// If the side is not touching the bounder then don't extend.
const int et_y = srcy ? 0 : dst->border;
const int el_y = srcx ? 0 : dst->border;
const int eb_y = srcy + srch != src->y_height ? 0 :
dst->border + dst->y_height - src->y_height;
const int er_y = srcx + srcw != src->y_width ? 0 :
dst->border + dst->y_width - src->y_width;
const int eb_y = srcy + srch != src->y_height
? 0
: dst->border + dst->y_height - src->y_height;
const int er_y = srcx + srcw != src->y_width
? 0
: dst->border + dst->y_width - src->y_width;
const int src_y_offset = srcy * src->y_stride + srcx;
const int dst_y_offset = srcy * dst->y_stride + srcx;
@@ -185,17 +178,14 @@ void vp9_copy_and_extend_frame_with_rect(const YV12_BUFFER_CONFIG *src,
const int srcw_uv = ROUND_POWER_OF_TWO(srcw, 1);
copy_and_extend_plane(src->y_buffer + src_y_offset, src->y_stride,
dst->y_buffer + dst_y_offset, dst->y_stride,
srcw, srch,
dst->y_buffer + dst_y_offset, dst->y_stride, srcw, srch,
et_y, el_y, eb_y, er_y);
copy_and_extend_plane(src->u_buffer + src_uv_offset, src->uv_stride,
dst->u_buffer + dst_uv_offset, dst->uv_stride,
srcw_uv, srch_uv,
et_uv, el_uv, eb_uv, er_uv);
dst->u_buffer + dst_uv_offset, dst->uv_stride, srcw_uv,
srch_uv, et_uv, el_uv, eb_uv, er_uv);
copy_and_extend_plane(src->v_buffer + src_uv_offset, src->uv_stride,
dst->v_buffer + dst_uv_offset, dst->uv_stride,
srcw_uv, srch_uv,
et_uv, el_uv, eb_uv, er_uv);
dst->v_buffer + dst_uv_offset, dst->uv_stride, srcw_uv,
srch_uv, et_uv, el_uv, eb_uv, er_uv);
}

6
vp9/encoder/vp9_extend.h
View File

@@ -18,14 +18,12 @@
extern "C" {
#endif
void vp9_copy_and_extend_frame(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst);
void vp9_copy_and_extend_frame_with_rect(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst,
int srcy, int srcx,
int srch, int srcw);
YV12_BUFFER_CONFIG *dst, int srcy,
int srcx, int srch, int srcw);
#ifdef __cplusplus
} // extern "C"
#endif

1027
vp9/encoder/vp9_firstpass.c
View File

File diff suppressed because it is too large Load Diff

3
vp9/encoder/vp9_firstpass.h
View File

@@ -154,8 +154,7 @@ void vp9_twopass_postencode_update(struct VP9_COMP *cpi);
// Post encode update of the rate control parameters for 2-pass
void vp9_twopass_postencode_update(struct VP9_COMP *cpi);
void calculate_coded_size(struct VP9_COMP *cpi,
int *scaled_frame_width,
void calculate_coded_size(struct VP9_COMP *cpi, int *scaled_frame_width,
int *scaled_frame_height);
#ifdef __cplusplus

67
vp9/encoder/vp9_lookahead.c
View File

@@ -19,33 +19,28 @@
#include "vp9/encoder/vp9_lookahead.h"
/* Return the buffer at the given absolute index and increment the index */
static struct lookahead_entry *pop(struct lookahead_ctx *ctx,
int *idx) {
static struct lookahead_entry *pop(struct lookahead_ctx *ctx, int *idx) {
int index = *idx;
struct lookahead_entry *buf = ctx->buf + index;
assert(index < ctx->max_sz);
if (++index >= ctx->max_sz)
index -= ctx->max_sz;
if (++index >= ctx->max_sz) index -= ctx->max_sz;
*idx = index;
return buf;
}
void vp9_lookahead_destroy(struct lookahead_ctx *ctx) {
if (ctx) {
if (ctx->buf) {
int i;
for (i = 0; i < ctx->max_sz; i++)
vpx_free_frame_buffer(&ctx->buf[i].img);
for (i = 0; i < ctx->max_sz; i++) vpx_free_frame_buffer(&ctx->buf[i].img);
free(ctx->buf);
}
free(ctx);
}
}
struct lookahead_ctx *vp9_lookahead_init(unsigned int width,
unsigned int height,
unsigned int subsampling_x,
@@ -69,20 +64,18 @@ struct lookahead_ctx *vp9_lookahead_init(unsigned int width,
unsigned int i;
ctx->max_sz = depth;
ctx->buf = calloc(depth, sizeof(*ctx->buf));
if (!ctx->buf)
goto bail;
if (!ctx->buf) goto bail;
for (i = 0; i < depth; i++)
if (vpx_alloc_frame_buffer(&ctx->buf[i].img,
width, height, subsampling_x, subsampling_y,
if (vpx_alloc_frame_buffer(
&ctx->buf[i].img, width, height, subsampling_x, subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
use_highbitdepth,
use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS,
legacy_byte_alignment))
VP9_ENC_BORDER_IN_PIXELS, legacy_byte_alignment))
goto bail;
}
return ctx;
bail:
bail:
vp9_lookahead_destroy(ctx);
return NULL;
}
@@ -109,8 +102,7 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
int subsampling_y = src->subsampling_y;
int larger_dimensions, new_dimensions;
if (ctx->sz + 1 + MAX_PRE_FRAMES > ctx->max_sz)
return 1;
if (ctx->sz + 1 + MAX_PRE_FRAMES > ctx->max_sz) return 1;
ctx->sz++;
buf = pop(ctx, &ctx->write_idx);
@@ -118,8 +110,7 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
height != buf->img.y_crop_height ||
uv_width != buf->img.uv_crop_width ||
uv_height != buf->img.uv_crop_height;
larger_dimensions = width > buf->img.y_width ||
height > buf->img.y_height ||
larger_dimensions = width > buf->img.y_width || height > buf->img.y_height ||
uv_width > buf->img.uv_width ||
uv_height > buf->img.uv_height;
assert(!larger_dimensions || new_dimensions);
@@ -139,27 +130,22 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
while (1) {
// Find the first active macroblock in this row.
for (; col < mb_cols; ++col) {
if (active_map[col])
break;
if (active_map[col]) break;
}
// No more active macroblock in this row.
if (col == mb_cols)
break;
if (col == mb_cols) break;
// Find the end of active region in this row.
active_end = col;
for (; active_end < mb_cols; ++active_end) {
if (!active_map[active_end])
break;
if (!active_map[active_end]) break;
}
// Only copy this active region.
vp9_copy_and_extend_frame_with_rect(src, &buf->img,
row << 4,
col << 4, 16,
(active_end - col) << 4);
vp9_copy_and_extend_frame_with_rect(src, &buf->img, row << 4, col << 4,
16, (active_end - col) << 4);
// Start again from the end of this active region.
col = active_end;
@@ -172,14 +158,13 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
if (larger_dimensions) {
YV12_BUFFER_CONFIG new_img;
memset(&new_img, 0, sizeof(new_img));
if (vpx_alloc_frame_buffer(&new_img,
width, height, subsampling_x, subsampling_y,
if (vpx_alloc_frame_buffer(&new_img, width, height, subsampling_x,
subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS,
0))
return 1;
VP9_ENC_BORDER_IN_PIXELS, 0))
return 1;
vpx_free_frame_buffer(&buf->img);
buf->img = new_img;
} else if (new_dimensions) {
@@ -202,7 +187,6 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
return 0;
}
struct lookahead_entry *vp9_lookahead_pop(struct lookahead_ctx *ctx,
int drain) {
struct lookahead_entry *buf = NULL;
@@ -214,7 +198,6 @@ struct lookahead_entry *vp9_lookahead_pop(struct lookahead_ctx *ctx,
return buf;
}
struct lookahead_entry *vp9_lookahead_peek(struct lookahead_ctx *ctx,
int index) {
struct lookahead_entry *buf = NULL;
@@ -223,16 +206,14 @@ struct lookahead_entry *vp9_lookahead_peek(struct lookahead_ctx *ctx,
// Forward peek
if (index < ctx->sz) {
index += ctx->read_idx;
if (index >= ctx->max_sz)
index -= ctx->max_sz;
if (index >= ctx->max_sz) index -= ctx->max_sz;
buf = ctx->buf + index;
}
} else if (index < 0) {
// Backward peek
if (-index <= MAX_PRE_FRAMES) {
index += ctx->read_idx;
if (index < 0)
index += ctx->max_sz;
if (index < 0) index += ctx->max_sz;
buf = ctx->buf + index;
}
}
@@ -240,6 +221,4 @@ struct lookahead_entry *vp9_lookahead_peek(struct lookahead_ctx *ctx,
return buf;
}
unsigned int vp9_lookahead_depth(struct lookahead_ctx *ctx) {
return ctx->sz;
}
unsigned int vp9_lookahead_depth(struct lookahead_ctx *ctx) { return ctx->sz; }

16
vp9/encoder/vp9_lookahead.h
View File

@@ -26,10 +26,10 @@ extern "C" {
#define MAX_LAG_BUFFERS 25
struct lookahead_entry {
YV12_BUFFER_CONFIG img;
int64_t ts_start;
int64_t ts_end;
unsigned int flags;
YV12_BUFFER_CONFIG img;
int64_t ts_start;
int64_t ts_end;
unsigned int flags;
};
// The max of past frames we want to keep in the queue.
@@ -57,12 +57,10 @@ struct lookahead_ctx *vp9_lookahead_init(unsigned int width,
#endif
unsigned int depth);
/**\brief Destroys the lookahead stage
*/
void vp9_lookahead_destroy(struct lookahead_ctx *ctx);
/**\brief Enqueue a source buffer
*
* This function will copy the source image into a new framebuffer with
@@ -85,7 +83,6 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
#endif
unsigned int flags);
/**\brief Get the next source buffer to encode
*
*
@@ -96,9 +93,7 @@ int vp9_lookahead_push(struct lookahead_ctx *ctx, YV12_BUFFER_CONFIG *src,
* \retval NULL, if drain set and queue is empty
* \retval NULL, if drain not set and queue not of the configured depth
*/
struct lookahead_entry *vp9_lookahead_pop(struct lookahead_ctx *ctx,
int drain);
struct lookahead_entry *vp9_lookahead_pop(struct lookahead_ctx *ctx, int drain);
/**\brief Get a future source buffer to encode
*
@@ -110,7 +105,6 @@ struct lookahead_entry *vp9_lookahead_pop(struct lookahead_ctx *ctx,
struct lookahead_entry *vp9_lookahead_peek(struct lookahead_ctx *ctx,
int index);
/**\brief Get the number of frames currently in the lookahead queue
*
* \param[in] ctx Pointer to the lookahead context

132
vp9/encoder/vp9_mbgraph.c
View File

@@ -22,11 +22,8 @@
#include "vp9/common/vp9_reconinter.h"
#include "vp9/common/vp9_reconintra.h"
static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi,
const MV *ref_mv,
MV *dst_mv,
int mb_row,
static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi, const MV *ref_mv,
MV *dst_mv, int mb_row,
int mb_col) {
MACROBLOCK *const x = &cpi->td.mb;
MACROBLOCKD *const xd = &x->e_mbd;
@@ -64,9 +61,8 @@ static unsigned int do_16x16_motion_iteration(VP9_COMP *cpi,
cpi->find_fractional_mv_step(
x, dst_mv, ref_mv, cpi->common.allow_high_precision_mv, x->errorperbit,
&v_fn_ptr, 0, mv_sf->subpel_iters_per_step,
cond_cost_list(cpi, cost_list),
NULL, NULL,
&distortion, &sse, NULL, 0, 0);
cond_cost_list(cpi, cost_list), NULL, NULL, &distortion, &sse, NULL, 0,
0);
}
xd->mi[0]->mode = NEWMV;
@@ -109,10 +105,10 @@ static int do_16x16_motion_search(VP9_COMP *cpi, const MV *ref_mv,
// based search as well.
if (ref_mv->row != 0 || ref_mv->col != 0) {
unsigned int tmp_err;
MV zero_ref_mv = {0, 0}, tmp_mv;
MV zero_ref_mv = { 0, 0 }, tmp_mv;
tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv, &tmp_mv,
mb_row, mb_col);
tmp_err =
do_16x16_motion_iteration(cpi, &zero_ref_mv, &tmp_mv, mb_row, mb_col);
if (tmp_err < err) {
dst_mv->as_mv = tmp_mv;
err = tmp_err;
@@ -137,7 +133,7 @@ static int do_16x16_zerozero_search(VP9_COMP *cpi, int_mv *dst_mv) {
return err;
}
static int find_best_16x16_intra(VP9_COMP *cpi, PREDICTION_MODE *pbest_mode) {
MACROBLOCK *const x = &cpi->td.mb;
MACROBLOCK *const x = &cpi->td.mb;
MACROBLOCKD *const xd = &x->e_mbd;
PREDICTION_MODE best_mode = -1, mode;
unsigned int best_err = INT_MAX;
@@ -148,38 +144,30 @@ static int find_best_16x16_intra(VP9_COMP *cpi, PREDICTION_MODE *pbest_mode) {
unsigned int err;
xd->mi[0]->mode = mode;
vp9_predict_intra_block(xd, 2, TX_16X16, mode,
x->plane[0].src.buf, x->plane[0].src.stride,
xd->plane[0].dst.buf, xd->plane[0].dst.stride,
0, 0, 0);
vp9_predict_intra_block(xd, 2, TX_16X16, mode, x->plane[0].src.buf,
x->plane[0].src.stride, xd->plane[0].dst.buf,
xd->plane[0].dst.stride, 0, 0, 0);
err = vpx_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
xd->plane[0].dst.buf, xd->plane[0].dst.stride);
// find best
if (err < best_err) {
best_err = err;
best_err = err;
best_mode = mode;
}
}
if (pbest_mode)
*pbest_mode = best_mode;
if (pbest_mode) *pbest_mode = best_mode;
return best_err;
}
static void update_mbgraph_mb_stats
(
VP9_COMP *cpi,
MBGRAPH_MB_STATS *stats,
YV12_BUFFER_CONFIG *buf,
int mb_y_offset,
YV12_BUFFER_CONFIG *golden_ref,
const MV *prev_golden_ref_mv,
YV12_BUFFER_CONFIG *alt_ref,
int mb_row,
int mb_col
) {
static void update_mbgraph_mb_stats(VP9_COMP *cpi, MBGRAPH_MB_STATS *stats,
YV12_BUFFER_CONFIG *buf, int mb_y_offset,
YV12_BUFFER_CONFIG *golden_ref,
const MV *prev_golden_ref_mv,
YV12_BUFFER_CONFIG *alt_ref, int mb_row,
int mb_col) {
MACROBLOCK *const x = &cpi->td.mb;
MACROBLOCKD *const xd = &x->e_mbd;
int intra_error;
@@ -193,10 +181,8 @@ static void update_mbgraph_mb_stats
xd->plane[0].dst.stride = get_frame_new_buffer(cm)->y_stride;
// do intra 16x16 prediction
intra_error = find_best_16x16_intra(cpi,
&stats->ref[INTRA_FRAME].m.mode);
if (intra_error <= 0)
intra_error = 1;
intra_error = find_best_16x16_intra(cpi, &stats->ref[INTRA_FRAME].m.mode);
if (intra_error <= 0) intra_error = 1;
stats->ref[INTRA_FRAME].err = intra_error;
// Golden frame MV search, if it exists and is different than last frame
@@ -204,10 +190,9 @@ static void update_mbgraph_mb_stats
int g_motion_error;
xd->plane[0].pre[0].buf = golden_ref->y_buffer + mb_y_offset;
xd->plane[0].pre[0].stride = golden_ref->y_stride;
g_motion_error = do_16x16_motion_search(cpi,
prev_golden_ref_mv,
&stats->ref[GOLDEN_FRAME].m.mv,
mb_row, mb_col);
g_motion_error =
do_16x16_motion_search(cpi, prev_golden_ref_mv,
&stats->ref[GOLDEN_FRAME].m.mv, mb_row, mb_col);
stats->ref[GOLDEN_FRAME].err = g_motion_error;
} else {
stats->ref[GOLDEN_FRAME].err = INT_MAX;
@@ -220,8 +205,8 @@ static void update_mbgraph_mb_stats
int a_motion_error;
xd->plane[0].pre[0].buf = alt_ref->y_buffer + mb_y_offset;
xd->plane[0].pre[0].stride = alt_ref->y_stride;
a_motion_error = do_16x16_zerozero_search(cpi,
&stats->ref[ALTREF_FRAME].m.mv);
a_motion_error =
do_16x16_zerozero_search(cpi, &stats->ref[ALTREF_FRAME].m.mv);
stats->ref[ALTREF_FRAME].err = a_motion_error;
} else {
@@ -241,20 +226,20 @@ static void update_mbgraph_frame_stats(VP9_COMP *cpi,
int mb_col, mb_row, offset = 0;
int mb_y_offset = 0, arf_y_offset = 0, gld_y_offset = 0;
MV gld_top_mv = {0, 0};
MV gld_top_mv = { 0, 0 };
MODE_INFO mi_local;
MODE_INFO mi_above, mi_left;
vp9_zero(mi_local);
// Set up limit values for motion vectors to prevent them extending outside
// the UMV borders.
x->mv_row_min = -BORDER_MV_PIXELS_B16;
x->mv_row_max = (cm->mb_rows - 1) * 8 + BORDER_MV_PIXELS_B16;
x->mv_row_min = -BORDER_MV_PIXELS_B16;
x->mv_row_max = (cm->mb_rows - 1) * 8 + BORDER_MV_PIXELS_B16;
// Signal to vp9_predict_intra_block() that above is not available
xd->above_mi = NULL;
xd->plane[0].dst.stride = buf->y_stride;
xd->plane[0].pre[0].stride = buf->y_stride;
xd->plane[0].dst.stride = buf->y_stride;
xd->plane[0].pre[0].stride = buf->y_stride;
xd->plane[1].dst.stride = buf->uv_stride;
xd->mi[0] = &mi_local;
mi_local.sb_type = BLOCK_16X16;
@@ -263,23 +248,22 @@ static void update_mbgraph_frame_stats(VP9_COMP *cpi,
for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
MV gld_left_mv = gld_top_mv;
int mb_y_in_offset = mb_y_offset;
int mb_y_in_offset = mb_y_offset;
int arf_y_in_offset = arf_y_offset;
int gld_y_in_offset = gld_y_offset;
// Set up limit values for motion vectors to prevent them extending outside
// the UMV borders.
x->mv_col_min = -BORDER_MV_PIXELS_B16;
x->mv_col_max = (cm->mb_cols - 1) * 8 + BORDER_MV_PIXELS_B16;
x->mv_col_min = -BORDER_MV_PIXELS_B16;
x->mv_col_max = (cm->mb_cols - 1) * 8 + BORDER_MV_PIXELS_B16;
// Signal to vp9_predict_intra_block() that left is not available
xd->left_mi = NULL;
for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
MBGRAPH_MB_STATS *mb_stats = &stats->mb_stats[offset + mb_col];
update_mbgraph_mb_stats(cpi, mb_stats, buf, mb_y_in_offset,
golden_ref, &gld_left_mv, alt_ref,
mb_row, mb_col);
update_mbgraph_mb_stats(cpi, mb_stats, buf, mb_y_in_offset, golden_ref,
&gld_left_mv, alt_ref, mb_row, mb_col);
gld_left_mv = mb_stats->ref[GOLDEN_FRAME].m.mv.as_mv;
if (mb_col == 0) {
gld_top_mv = gld_left_mv;
@@ -287,23 +271,22 @@ static void update_mbgraph_frame_stats(VP9_COMP *cpi,
// Signal to vp9_predict_intra_block() that left is available
xd->left_mi = &mi_left;
mb_y_in_offset += 16;
gld_y_in_offset += 16;
arf_y_in_offset += 16;
x->mv_col_min -= 16;
x->mv_col_max -= 16;
mb_y_in_offset += 16;
gld_y_in_offset += 16;
arf_y_in_offset += 16;
x->mv_col_min -= 16;
x->mv_col_max -= 16;
}
// Signal to vp9_predict_intra_block() that above is available
xd->above_mi = &mi_above;
mb_y_offset += buf->y_stride * 16;
gld_y_offset += golden_ref->y_stride * 16;
if (alt_ref)
arf_y_offset += alt_ref->y_stride * 16;
x->mv_row_min -= 16;
x->mv_row_max -= 16;
offset += cm->mb_cols;
mb_y_offset += buf->y_stride * 16;
gld_y_offset += golden_ref->y_stride * 16;
if (alt_ref) arf_y_offset += alt_ref->y_stride * 16;
x->mv_row_min -= 16;
x->mv_row_max -= 16;
offset += cm->mb_cols;
}
}
@@ -317,9 +300,9 @@ static void separate_arf_mbs(VP9_COMP *cpi) {
int *arf_not_zz;
CHECK_MEM_ERROR(cm, arf_not_zz,
vpx_calloc(cm->mb_rows * cm->mb_cols * sizeof(*arf_not_zz),
1));
CHECK_MEM_ERROR(
cm, arf_not_zz,
vpx_calloc(cm->mb_rows * cm->mb_cols * sizeof(*arf_not_zz), 1));
// We are not interested in results beyond the alt ref itself.
if (n_frames > cpi->rc.frames_till_gf_update_due)
@@ -335,12 +318,11 @@ static void separate_arf_mbs(VP9_COMP *cpi) {
MBGRAPH_MB_STATS *mb_stats = &frame_stats->mb_stats[offset + mb_col];
int altref_err = mb_stats->ref[ALTREF_FRAME].err;
int intra_err = mb_stats->ref[INTRA_FRAME ].err;
int intra_err = mb_stats->ref[INTRA_FRAME].err;
int golden_err = mb_stats->ref[GOLDEN_FRAME].err;
// Test for altref vs intra and gf and that its mv was 0,0.
if (altref_err > 1000 ||
altref_err > intra_err ||
if (altref_err > 1000 || altref_err > intra_err ||
altref_err > golden_err) {
arf_not_zz[offset + mb_col]++;
}
@@ -395,11 +377,9 @@ void vp9_update_mbgraph_stats(VP9_COMP *cpi) {
// we need to look ahead beyond where the ARF transitions into
// being a GF - so exit if we don't look ahead beyond that
if (n_frames <= cpi->rc.frames_till_gf_update_due)
return;
if (n_frames <= cpi->rc.frames_till_gf_update_due) return;
if (n_frames > MAX_LAG_BUFFERS)
n_frames = MAX_LAG_BUFFERS;
if (n_frames > MAX_LAG_BUFFERS) n_frames = MAX_LAG_BUFFERS;
cpi->mbgraph_n_frames = n_frames;
for (i = 0; i < n_frames; i++) {
@@ -418,8 +398,8 @@ void vp9_update_mbgraph_stats(VP9_COMP *cpi) {
assert(q_cur != NULL);
update_mbgraph_frame_stats(cpi, frame_stats, &q_cur->img,
golden_ref, cpi->Source);
update_mbgraph_frame_stats(cpi, frame_stats, &q_cur->img, golden_ref,
cpi->Source);
}
vpx_clear_system_state();

4
vp9/encoder/vp9_mbgraph.h
View File

@@ -25,9 +25,7 @@ typedef struct {
} ref[MAX_REF_FRAMES];
} MBGRAPH_MB_STATS;
typedef struct {
MBGRAPH_MB_STATS *mb_stats;
} MBGRAPH_FRAME_STATS;
typedef struct { MBGRAPH_MB_STATS *mb_stats; } MBGRAPH_FRAME_STATS;
struct VP9_COMP;

1552
vp9/encoder/vp9_mcomp.c
View File

File diff suppressed because it is too large Load Diff

82
vp9/encoder/vp9_mcomp.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_MCOMP_H_
#define VP9_ENCODER_VP9_MCOMP_H_
@@ -26,67 +25,56 @@ extern "C" {
// Enable the use of motion vector in range [-1023, 1023].
#define MAX_FULL_PEL_VAL ((1 << (MAX_MVSEARCH_STEPS - 1)) - 1)
// Maximum size of the first step in full pel units
#define MAX_FIRST_STEP (1 << (MAX_MVSEARCH_STEPS-1))
#define MAX_FIRST_STEP (1 << (MAX_MVSEARCH_STEPS - 1))
// Allowed motion vector pixel distance outside image border
// for Block_16x16
#define BORDER_MV_PIXELS_B16 (16 + VP9_INTERP_EXTEND)
typedef struct search_site_config {
// motion search sites
MV ss_mv[8 * MAX_MVSEARCH_STEPS]; // Motion vector
intptr_t ss_os[8 * MAX_MVSEARCH_STEPS]; // Offset
MV ss_mv[8 * MAX_MVSEARCH_STEPS]; // Motion vector
intptr_t ss_os[8 * MAX_MVSEARCH_STEPS]; // Offset
int searches_per_step;
int total_steps;
} search_site_config;
void vp9_init_dsmotion_compensation(search_site_config *cfg, int stride);
void vp9_init3smotion_compensation(search_site_config *cfg, int stride);
void vp9_init3smotion_compensation(search_site_config *cfg, int stride);
void vp9_set_mv_search_range(MACROBLOCK *x, const MV *mv);
int vp9_mv_bit_cost(const MV *mv, const MV *ref,
const int *mvjcost, int *mvcost[2], int weight);
int vp9_mv_bit_cost(const MV *mv, const MV *ref, const int *mvjcost,
int *mvcost[2], int weight);
// Utility to compute variance + MV rate cost for a given MV
int vp9_get_mvpred_var(const MACROBLOCK *x,
const MV *best_mv, const MV *center_mv,
const vp9_variance_fn_ptr_t *vfp,
int vp9_get_mvpred_var(const MACROBLOCK *x, const MV *best_mv,
const MV *center_mv, const vp9_variance_fn_ptr_t *vfp,
int use_mvcost);
int vp9_get_mvpred_av_var(const MACROBLOCK *x,
const MV *best_mv, const MV *center_mv,
const uint8_t *second_pred,
const vp9_variance_fn_ptr_t *vfp,
int use_mvcost);
int vp9_get_mvpred_av_var(const MACROBLOCK *x, const MV *best_mv,
const MV *center_mv, const uint8_t *second_pred,
const vp9_variance_fn_ptr_t *vfp, int use_mvcost);
struct VP9_COMP;
struct SPEED_FEATURES;
int vp9_init_search_range(int size);
int vp9_refining_search_sad(const struct macroblock *x,
struct mv *ref_mv,
int vp9_refining_search_sad(const struct macroblock *x, struct mv *ref_mv,
int sad_per_bit, int distance,
const struct vp9_variance_vtable *fn_ptr,
const struct mv *center_mv);
// Perform integral projection based motion estimation.
unsigned int vp9_int_pro_motion_estimation(const struct VP9_COMP *cpi,
MACROBLOCK *x,
BLOCK_SIZE bsize,
MACROBLOCK *x, BLOCK_SIZE bsize,
int mi_row, int mi_col);
typedef uint32_t (fractional_mv_step_fp) (
const MACROBLOCK *x,
MV *bestmv, const MV *ref_mv,
int allow_hp,
int error_per_bit,
const vp9_variance_fn_ptr_t *vfp,
typedef uint32_t(fractional_mv_step_fp)(
const MACROBLOCK *x, MV *bestmv, const MV *ref_mv, int allow_hp,
int error_per_bit, const vp9_variance_fn_ptr_t *vfp,
int forced_stop, // 0 - full, 1 - qtr only, 2 - half only
int iters_per_step,
int *cost_list,
int *mvjcost, int *mvcost[2],
uint32_t *distortion, uint32_t *sse1,
const uint8_t *second_pred,
int w, int h);
int iters_per_step, int *cost_list, int *mvjcost, int *mvcost[2],
uint32_t *distortion, uint32_t *sse1, const uint8_t *second_pred, int w,
int h);
extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree;
extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned;
@@ -94,39 +82,31 @@ extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned_more;
extern fractional_mv_step_fp vp9_find_best_sub_pixel_tree_pruned_evenmore;
extern fractional_mv_step_fp vp9_skip_sub_pixel_tree;
typedef int (*vp9_full_search_fn_t)(const MACROBLOCK *x,
const MV *ref_mv, int sad_per_bit,
int distance,
typedef int (*vp9_full_search_fn_t)(const MACROBLOCK *x, const MV *ref_mv,
int sad_per_bit, int distance,
const vp9_variance_fn_ptr_t *fn_ptr,
const MV *center_mv, MV *best_mv);
typedef int (*vp9_refining_search_fn_t)(const MACROBLOCK *x,
MV *ref_mv, int sad_per_bit,
int distance,
typedef int (*vp9_refining_search_fn_t)(const MACROBLOCK *x, MV *ref_mv,
int sad_per_bit, int distance,
const vp9_variance_fn_ptr_t *fn_ptr,
const MV *center_mv);
typedef int (*vp9_diamond_search_fn_t)(const MACROBLOCK *x,
const search_site_config *cfg,
MV *ref_mv, MV *best_mv,
int search_param, int sad_per_bit,
int *num00,
const vp9_variance_fn_ptr_t *fn_ptr,
const MV *center_mv);
typedef int (*vp9_diamond_search_fn_t)(
const MACROBLOCK *x, const search_site_config *cfg, MV *ref_mv, MV *best_mv,
int search_param, int sad_per_bit, int *num00,
const vp9_variance_fn_ptr_t *fn_ptr, const MV *center_mv);
int vp9_refining_search_8p_c(const MACROBLOCK *x,
MV *ref_mv, int error_per_bit,
int vp9_refining_search_8p_c(const MACROBLOCK *x, MV *ref_mv, int error_per_bit,
int search_range,
const vp9_variance_fn_ptr_t *fn_ptr,
const MV *center_mv, const uint8_t *second_pred);
struct VP9_COMP;
int vp9_full_pixel_search(struct VP9_COMP *cpi, MACROBLOCK *x,
BLOCK_SIZE bsize, MV *mvp_full,
int step_param, int error_per_bit,
int *cost_list,
const MV *ref_mv, MV *tmp_mv,
int vp9_full_pixel_search(struct VP9_COMP *cpi, MACROBLOCK *x, BLOCK_SIZE bsize,
MV *mvp_full, int step_param, int error_per_bit,
int *cost_list, const MV *ref_mv, MV *tmp_mv,
int var_max, int rd);
#ifdef __cplusplus

78
vp9/encoder/vp9_noise_estimate.c
View File

@@ -21,9 +21,7 @@
#include "vp9/encoder/vp9_noise_estimate.h"
#include "vp9/encoder/vp9_encoder.h"
void vp9_noise_estimate_init(NOISE_ESTIMATE *const ne,
int width,
int height) {
void vp9_noise_estimate_init(NOISE_ESTIMATE *const ne, int width, int height) {
ne->enabled = 0;
ne->level = kLowLow;
ne->value = 0;
@@ -40,24 +38,18 @@ void vp9_noise_estimate_init(NOISE_ESTIMATE *const ne,
}
static int enable_noise_estimation(VP9_COMP *const cpi) {
// Enable noise estimation if denoising is on.
// Enable noise estimation if denoising is on.
#if CONFIG_VP9_TEMPORAL_DENOISING
if (cpi->oxcf.noise_sensitivity > 0)
return 1;
if (cpi->oxcf.noise_sensitivity > 0) return 1;
#endif
// Only allow noise estimate under certain encoding mode.
// Enabled for 1 pass CBR, speed >=5, and if resolution is same as original.
// Not enabled for SVC mode and screen_content_mode.
// Not enabled for low resolutions.
if (cpi->oxcf.pass == 0 &&
cpi->oxcf.rc_mode == VPX_CBR &&
cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
cpi->oxcf.speed >= 5 &&
cpi->resize_state == ORIG &&
cpi->resize_pending == 0 &&
!cpi->use_svc &&
cpi->oxcf.content != VP9E_CONTENT_SCREEN &&
cpi->common.width >= 640 &&
if (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == VPX_CBR &&
cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cpi->oxcf.speed >= 5 &&
cpi->resize_state == ORIG && cpi->resize_pending == 0 && !cpi->use_svc &&
cpi->oxcf.content != VP9E_CONTENT_SCREEN && cpi->common.width >= 640 &&
cpi->common.height >= 480)
return 1;
else
@@ -65,8 +57,8 @@ static int enable_noise_estimation(VP9_COMP *const cpi) {
}
#if CONFIG_VP9_TEMPORAL_DENOISING
static void copy_frame(YV12_BUFFER_CONFIG * const dest,
const YV12_BUFFER_CONFIG * const src) {
static void copy_frame(YV12_BUFFER_CONFIG *const dest,
const YV12_BUFFER_CONFIG *const src) {
int r;
const uint8_t *srcbuf = src->y_buffer;
uint8_t *destbuf = dest->y_buffer;
@@ -110,18 +102,15 @@ void vp9_update_noise_estimate(VP9_COMP *const cpi) {
// Estimate is between current source and last source.
YV12_BUFFER_CONFIG *last_source = cpi->Last_Source;
#if CONFIG_VP9_TEMPORAL_DENOISING
if (cpi->oxcf.noise_sensitivity > 0)
last_source = &cpi->denoiser.last_source;
if (cpi->oxcf.noise_sensitivity > 0) last_source = &cpi->denoiser.last_source;
#endif
ne->enabled = enable_noise_estimation(cpi);
if (!ne->enabled ||
cm->current_video_frame % frame_period != 0 ||
last_source == NULL ||
ne->last_w != cm->width ||
if (!ne->enabled || cm->current_video_frame % frame_period != 0 ||
last_source == NULL || ne->last_w != cm->width ||
ne->last_h != cm->height) {
#if CONFIG_VP9_TEMPORAL_DENOISING
if (cpi->oxcf.noise_sensitivity > 0)
copy_frame(&cpi->denoiser.last_source, cpi->Source);
if (cpi->oxcf.noise_sensitivity > 0)
copy_frame(&cpi->denoiser.last_source, cpi->Source);
#endif
if (last_source != NULL) {
ne->last_w = cm->width;
@@ -140,8 +129,8 @@ void vp9_update_noise_estimate(VP9_COMP *const cpi) {
int num_samples = 0;
uint64_t avg_est = 0;
int bsize = BLOCK_16X16;
static const unsigned char const_source[16] = {
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
static const unsigned char const_source[16] = { 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0 };
// Loop over sub-sample of 16x16 blocks of frame, and for blocks that have
// been encoded as zero/small mv at least x consecutive frames, compute
// the variance to update estimate of noise in the source.
@@ -167,8 +156,7 @@ void vp9_update_noise_estimate(VP9_COMP *const cpi) {
for (mi_row = 0; mi_row < cm->mi_rows; mi_row++) {
for (mi_col = 0; mi_col < cm->mi_cols; mi_col++) {
// 16x16 blocks, 1/4 sample of frame.
if (mi_row % 4 == 0 && mi_col % 4 == 0 &&
mi_row < cm->mi_rows - 1 &&
if (mi_row % 4 == 0 && mi_col % 4 == 0 && mi_row < cm->mi_rows - 1 &&
mi_col < cm->mi_cols - 1) {
int bl_index = mi_row * cm->mi_cols + mi_col;
int bl_index1 = bl_index + 1;
@@ -178,20 +166,16 @@ void vp9_update_noise_estimate(VP9_COMP *const cpi) {
// been encoded as zero/low motion x (= thresh_consec_zeromv) frames
// in a row. consec_zero_mv[] defined for 8x8 blocks, so consider all
// 4 sub-blocks for 16x16 block. Also, avoid skin blocks.
int consec_zeromv = VPXMIN(cpi->consec_zero_mv[bl_index],
VPXMIN(cpi->consec_zero_mv[bl_index1],
VPXMIN(cpi->consec_zero_mv[bl_index2],
cpi->consec_zero_mv[bl_index3])));
int consec_zeromv =
VPXMIN(cpi->consec_zero_mv[bl_index],
VPXMIN(cpi->consec_zero_mv[bl_index1],
VPXMIN(cpi->consec_zero_mv[bl_index2],
cpi->consec_zero_mv[bl_index3])));
int is_skin = 0;
if (cpi->use_skin_detection) {
is_skin = vp9_compute_skin_block(src_y,
src_u,
src_v,
src_ystride,
src_uvstride,
bsize,
consec_zeromv,
0);
is_skin =
vp9_compute_skin_block(src_y, src_u, src_v, src_ystride,
src_uvstride, bsize, consec_zeromv, 0);
}
if (frame_low_motion &&
cpi->consec_zero_mv[bl_index] > thresh_consec_zeromv &&
@@ -201,19 +185,15 @@ void vp9_update_noise_estimate(VP9_COMP *const cpi) {
!is_skin) {
// Compute variance.
unsigned int sse;
unsigned int variance = cpi->fn_ptr[bsize].vf(src_y,
src_ystride,
last_src_y,
last_src_ystride,
&sse);
unsigned int variance = cpi->fn_ptr[bsize].vf(
src_y, src_ystride, last_src_y, last_src_ystride, &sse);
// Only consider this block as valid for noise measurement if the
// average term (sse - variance = N * avg^{2}, N = 16X16) of the
// temporal residual is small (avoid effects from lighting change).
if ((sse - variance) < thresh_sum_diff) {
unsigned int sse2;
const unsigned int spatial_variance =
cpi->fn_ptr[bsize].vf(src_y, src_ystride, const_source,
0, &sse2);
const unsigned int spatial_variance = cpi->fn_ptr[bsize].vf(
src_y, src_ystride, const_source, 0, &sse2);
// Avoid blocks with high brightness and high spatial variance.
if ((sse2 - spatial_variance) < thresh_sum_spatial &&
spatial_variance < thresh_spatial_var) {

11
vp9/encoder/vp9_noise_estimate.h
View File

@@ -23,12 +23,7 @@
extern "C" {
#endif
typedef enum noise_level {
kLowLow,
kLow,
kMedium,
kHigh
} NOISE_LEVEL;
typedef enum noise_level { kLowLow, kLow, kMedium, kHigh } NOISE_LEVEL;
typedef struct noise_estimate {
int enabled;
@@ -43,9 +38,7 @@ typedef struct noise_estimate {
struct VP9_COMP;
void vp9_noise_estimate_init(NOISE_ESTIMATE *const ne,
int width,
int height);
void vp9_noise_estimate_init(NOISE_ESTIMATE *const ne, int width, int height);
NOISE_LEVEL vp9_noise_estimate_extract_level(NOISE_ESTIMATE *const ne);

39
vp9/encoder/vp9_picklpf.c
View File

@@ -33,10 +33,9 @@ static int get_max_filter_level(const VP9_COMP *cpi) {
}
}
static int64_t try_filter_frame(const YV12_BUFFER_CONFIG *sd,
VP9_COMP *const cpi,
int filt_level, int partial_frame) {
VP9_COMP *const cpi, int filt_level,
int partial_frame) {
VP9_COMMON *const cm = &cpi->common;
int64_t filt_err;
@@ -44,8 +43,8 @@ static int64_t try_filter_frame(const YV12_BUFFER_CONFIG *sd,
if (cpi->num_workers > 1)
vp9_loop_filter_frame_mt(cm->frame_to_show, cm, cpi->td.mb.e_mbd.plane,
filt_level, 1, partial_frame,
cpi->workers, cpi->num_workers, &cpi->lf_row_sync);
filt_level, 1, partial_frame, cpi->workers,
cpi->num_workers, &cpi->lf_row_sync);
else
vp9_loop_filter_frame(cm->frame_to_show, cm, &cpi->td.mb.e_mbd, filt_level,
1, partial_frame);
@@ -78,8 +77,7 @@ static int search_filter_level(const YV12_BUFFER_CONFIG *sd, VP9_COMP *cpi,
// Start the search at the previous frame filter level unless it is now out of
// range.
int filt_mid =
clamp(lf->last_filt_level, min_filter_level, max_filter_level);
int filt_mid = clamp(lf->last_filt_level, min_filter_level, max_filter_level);
int filter_step = filt_mid < 16 ? 4 : filt_mid / 4;
// Sum squared error at each filter level
int64_t ss_err[MAX_LOOP_FILTER + 1];
@@ -105,8 +103,7 @@ static int search_filter_level(const YV12_BUFFER_CONFIG *sd, VP9_COMP *cpi,
bias = (bias * cpi->twopass.section_intra_rating) / 20;
// yx, bias less for large block size
if (cm->tx_mode != ONLY_4X4)
bias >>= 1;
if (cm->tx_mode != ONLY_4X4) bias >>= 1;
if (filt_direction <= 0 && filt_low != filt_mid) {
// Get Low filter error score
@@ -117,8 +114,7 @@ static int search_filter_level(const YV12_BUFFER_CONFIG *sd, VP9_COMP *cpi,
// filter value.
if ((ss_err[filt_low] - bias) < best_err) {
// Was it actually better than the previous best?
if (ss_err[filt_low] < best_err)
best_err = ss_err[filt_low];
if (ss_err[filt_low] < best_err) best_err = ss_err[filt_low];
filt_best = filt_low;
}
@@ -154,17 +150,16 @@ void vp9_pick_filter_level(const YV12_BUFFER_CONFIG *sd, VP9_COMP *cpi,
VP9_COMMON *const cm = &cpi->common;
struct loopfilter *const lf = &cm->lf;
lf->sharpness_level = cm->frame_type == KEY_FRAME ? 0
: cpi->oxcf.sharpness;
lf->sharpness_level = cm->frame_type == KEY_FRAME ? 0 : cpi->oxcf.sharpness;
if (method == LPF_PICK_MINIMAL_LPF && lf->filter_level) {
lf->filter_level = 0;
lf->filter_level = 0;
} else if (method >= LPF_PICK_FROM_Q) {
const int min_filter_level = 0;
const int max_filter_level = get_max_filter_level(cpi);
const int q = vp9_ac_quant(cm->base_qindex, 0, cm->bit_depth);
// These values were determined by linear fitting the result of the
// searched level, filt_guess = q * 0.316206 + 3.87252
// These values were determined by linear fitting the result of the
// searched level, filt_guess = q * 0.316206 + 3.87252
#if CONFIG_VP9_HIGHBITDEPTH
int filt_guess;
switch (cm->bit_depth) {
@@ -178,18 +173,18 @@ void vp9_pick_filter_level(const YV12_BUFFER_CONFIG *sd, VP9_COMP *cpi,
filt_guess = ROUND_POWER_OF_TWO(q * 20723 + 16242526, 22);
break;
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 "
"or VPX_BITS_12");
assert(0 &&
"bit_depth should be VPX_BITS_8, VPX_BITS_10 "
"or VPX_BITS_12");
return;
}
#else
int filt_guess = ROUND_POWER_OF_TWO(q * 20723 + 1015158, 18);
#endif // CONFIG_VP9_HIGHBITDEPTH
if (cm->frame_type == KEY_FRAME)
filt_guess -= 4;
if (cm->frame_type == KEY_FRAME) filt_guess -= 4;
lf->filter_level = clamp(filt_guess, min_filter_level, max_filter_level);
} else {
lf->filter_level = search_filter_level(sd, cpi,
method == LPF_PICK_FROM_SUBIMAGE);
lf->filter_level =
search_filter_level(sd, cpi, method == LPF_PICK_FROM_SUBIMAGE);
}
}

1
vp9/encoder/vp9_picklpf.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_PICKLPF_H_
#define VP9_ENCODER_VP9_PICKLPF_H_

756
vp9/encoder/vp9_pickmode.c
View File

File diff suppressed because it is too large Load Diff

11
vp9/encoder/vp9_pickmode.h
View File

@@ -20,15 +20,12 @@ extern "C" {
void vp9_pick_intra_mode(VP9_COMP *cpi, MACROBLOCK *x, RD_COST *rd_cost,
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx);
void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x,
TileDataEnc *tile_data,
void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data,
int mi_row, int mi_col, RD_COST *rd_cost,
BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx);
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx);
void vp9_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x,
int mi_row, int mi_col, RD_COST *rd_cost,
BLOCK_SIZE bsize,
void vp9_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x, int mi_row,
int mi_col, RD_COST *rd_cost, BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx);
#ifdef __cplusplus

125
vp9/encoder/vp9_quantize.c
View File

@@ -21,13 +21,12 @@
#include "vp9/encoder/vp9_rd.h"
void vp9_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block,
const int16_t *zbin_ptr, const int16_t *round_ptr,
const int16_t *quant_ptr, const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan,
const int16_t *iscan) {
int i, eob = -1;
// TODO(jingning) Decide the need of these arguments after the
// quantization process is completed.
@@ -53,27 +52,21 @@ void vp9_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
if (tmp)
eob = i;
if (tmp) eob = i;
}
}
*eob_ptr = eob + 1;
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_quantize_fp_c(const tran_low_t *coeff_ptr,
intptr_t count,
int skip_block,
const int16_t *zbin_ptr,
void vp9_highbd_quantize_fp_c(const tran_low_t *coeff_ptr, intptr_t count,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr,
const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *scan,
const int16_t *iscan) {
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
int i;
int eob = -1;
// TODO(jingning) Decide the need of these arguments after the
@@ -97,8 +90,7 @@ void vp9_highbd_quantize_fp_c(const tran_low_t *coeff_ptr,
const int abs_qcoeff = (int)((tmp * quant_ptr[rc != 0]) >> 16);
qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
if (abs_qcoeff)
eob = i;
if (abs_qcoeff) eob = i;
}
}
*eob_ptr = eob + 1;
@@ -108,13 +100,11 @@ void vp9_highbd_quantize_fp_c(const tran_low_t *coeff_ptr,
// TODO(jingning) Refactor this file and combine functions with similar
// operations.
void vp9_quantize_fp_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block,
const int16_t *zbin_ptr, const int16_t *round_ptr,
const int16_t *quant_ptr,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
int i, eob = -1;
(void)zbin_ptr;
@@ -140,25 +130,19 @@ void vp9_quantize_fp_32x32_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
}
if (tmp)
eob = i;
if (tmp) eob = i;
}
}
*eob_ptr = eob + 1;
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_quantize_fp_32x32_c(const tran_low_t *coeff_ptr,
intptr_t n_coeffs, int skip_block,
const int16_t *zbin_ptr,
const int16_t *round_ptr,
const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr,
uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
void vp9_highbd_quantize_fp_32x32_c(
const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block,
const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
int i, eob = -1;
(void)zbin_ptr;
(void)quant_shift_ptr;
@@ -176,15 +160,14 @@ void vp9_highbd_quantize_fp_32x32_c(const tran_low_t *coeff_ptr,
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
if (abs_coeff >= (dequant_ptr[rc != 0] >> 2)) {
const int64_t tmp = abs_coeff
+ ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
abs_qcoeff = (uint32_t) ((tmp * quant_ptr[rc != 0]) >> 15);
const int64_t tmp =
abs_coeff + ROUND_POWER_OF_TWO(round_ptr[rc != 0], 1);
abs_qcoeff = (uint32_t)((tmp * quant_ptr[rc != 0]) >> 15);
qcoeff_ptr[rc] = (tran_low_t)((abs_qcoeff ^ coeff_sign) - coeff_sign);
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0] / 2;
}
if (abs_qcoeff)
eob = i;
if (abs_qcoeff) eob = i;
}
}
*eob_ptr = eob + 1;
@@ -199,30 +182,26 @@ void vp9_regular_quantize_b_4x4(MACROBLOCK *x, int plane, int block,
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
vpx_highbd_quantize_b(BLOCK_OFFSET(p->coeff, block),
16, x->skip_block,
vpx_highbd_quantize_b(BLOCK_OFFSET(p->coeff, block), 16, x->skip_block,
p->zbin, p->round, p->quant, p->quant_shift,
BLOCK_OFFSET(p->qcoeff, block),
BLOCK_OFFSET(pd->dqcoeff, block),
pd->dequant, &p->eobs[block],
scan, iscan);
BLOCK_OFFSET(pd->dqcoeff, block), pd->dequant,
&p->eobs[block], scan, iscan);
return;
}
#endif
vpx_quantize_b(BLOCK_OFFSET(p->coeff, block),
16, x->skip_block,
p->zbin, p->round, p->quant, p->quant_shift,
vpx_quantize_b(BLOCK_OFFSET(p->coeff, block), 16, x->skip_block, p->zbin,
p->round, p->quant, p->quant_shift,
BLOCK_OFFSET(p->qcoeff, block),
BLOCK_OFFSET(pd->dqcoeff, block),
pd->dequant, &p->eobs[block], scan, iscan);
BLOCK_OFFSET(pd->dqcoeff, block), pd->dequant, &p->eobs[block],
scan, iscan);
}
static void invert_quant(int16_t *quant, int16_t *shift, int d) {
unsigned t;
int l, m;
t = d;
for (l = 0; t > 1; l++)
t >>= 1;
for (l = 0; t > 1; l++) t >>= 1;
m = 1 + (1 << (16 + l)) / d;
*quant = (int16_t)(m - (1 << 16));
*shift = 1 << (16 - l);
@@ -232,18 +211,15 @@ static int get_qzbin_factor(int q, vpx_bit_depth_t bit_depth) {
const int quant = vp9_dc_quant(q, 0, bit_depth);
#if CONFIG_VP9_HIGHBITDEPTH
switch (bit_depth) {
case VPX_BITS_8:
return q == 0 ? 64 : (quant < 148 ? 84 : 80);
case VPX_BITS_10:
return q == 0 ? 64 : (quant < 592 ? 84 : 80);
case VPX_BITS_12:
return q == 0 ? 64 : (quant < 2368 ? 84 : 80);
case VPX_BITS_8: return q == 0 ? 64 : (quant < 148 ? 84 : 80);
case VPX_BITS_10: return q == 0 ? 64 : (quant < 592 ? 84 : 80);
case VPX_BITS_12: return q == 0 ? 64 : (quant < 2368 ? 84 : 80);
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
return -1;
}
#else
(void) bit_depth;
(void)bit_depth;
return q == 0 ? 64 : (quant < 148 ? 84 : 80);
#endif
}
@@ -259,8 +235,7 @@ void vp9_init_quantizer(VP9_COMP *cpi) {
for (i = 0; i < 2; ++i) {
int qrounding_factor_fp = i == 0 ? 48 : 42;
if (q == 0)
qrounding_factor_fp = 64;
if (q == 0) qrounding_factor_fp = 64;
// y
quant = i == 0 ? vp9_dc_quant(q, cm->y_dc_delta_q, cm->bit_depth)
@@ -275,8 +250,8 @@ void vp9_init_quantizer(VP9_COMP *cpi) {
// uv
quant = i == 0 ? vp9_dc_quant(q, cm->uv_dc_delta_q, cm->bit_depth)
: vp9_ac_quant(q, cm->uv_ac_delta_q, cm->bit_depth);
invert_quant(&quants->uv_quant[q][i],
&quants->uv_quant_shift[q][i], quant);
invert_quant(&quants->uv_quant[q][i], &quants->uv_quant_shift[q][i],
quant);
quants->uv_quant_fp[q][i] = (1 << 16) / quant;
quants->uv_round_fp[q][i] = (qrounding_factor_fp * quant) >> 7;
quants->uv_zbin[q][i] = ROUND_POWER_OF_TWO(qzbin_factor * quant, 7);
@@ -363,14 +338,11 @@ void vp9_set_quantizer(VP9_COMMON *cm, int q) {
// Table that converts 0-63 Q-range values passed in outside to the Qindex
// range used internally.
static const int quantizer_to_qindex[] = {
0, 4, 8, 12, 16, 20, 24, 28,
32, 36, 40, 44, 48, 52, 56, 60,
64, 68, 72, 76, 80, 84, 88, 92,
96, 100, 104, 108, 112, 116, 120, 124,
128, 132, 136, 140, 144, 148, 152, 156,
160, 164, 168, 172, 176, 180, 184, 188,
192, 196, 200, 204, 208, 212, 216, 220,
224, 228, 232, 236, 240, 244, 249, 255,
0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48,
52, 56, 60, 64, 68, 72, 76, 80, 84, 88, 92, 96, 100,
104, 108, 112, 116, 120, 124, 128, 132, 136, 140, 144, 148, 152,
156, 160, 164, 168, 172, 176, 180, 184, 188, 192, 196, 200, 204,
208, 212, 216, 220, 224, 228, 232, 236, 240, 244, 249, 255,
};
int vp9_quantizer_to_qindex(int quantizer) {
@@ -381,8 +353,7 @@ int vp9_qindex_to_quantizer(int qindex) {
int quantizer;
for (quantizer = 0; quantizer < 64; ++quantizer)
if (quantizer_to_qindex[quantizer] >= qindex)
return quantizer;
if (quantizer_to_qindex[quantizer] >= qindex) return quantizer;
return 63;
}

722
vp9/encoder/vp9_ratectrl.c
View File

File diff suppressed because it is too large Load Diff

41
vp9/encoder/vp9_ratectrl.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_RATECTRL_H_
#define VP9_ENCODER_VP9_RATECTRL_H_
@@ -23,12 +22,12 @@ extern "C" {
#endif
// Bits Per MB at different Q (Multiplied by 512)
#define BPER_MB_NORMBITS 9
#define BPER_MB_NORMBITS 9
#define MIN_GF_INTERVAL 4
#define MAX_GF_INTERVAL 16
#define FIXED_GF_INTERVAL 8 // Used in some testing modes only
#define ONEHALFONLY_RESIZE 0
#define MIN_GF_INTERVAL 4
#define MAX_GF_INTERVAL 16
#define FIXED_GF_INTERVAL 8 // Used in some testing modes only
#define ONEHALFONLY_RESIZE 0
typedef enum {
INTER_NORMAL = 0,
@@ -54,36 +53,32 @@ typedef enum {
UP_ORIG = -2, // From 1/2 or 3/4 to orig.
} RESIZE_ACTION;
typedef enum {
ORIG = 0,
THREE_QUARTER = 1,
ONE_HALF = 2
} RESIZE_STATE;
typedef enum { ORIG = 0, THREE_QUARTER = 1, ONE_HALF = 2 } RESIZE_STATE;
// Frame dimensions multiplier wrt the native frame size, in 1/16ths,
// specified for the scale-up case.
// e.g. 24 => 16/24 = 2/3 of native size. The restriction to 1/16th is
// intended to match the capabilities of the normative scaling filters,
// giving precedence to the up-scaling accuracy.
static const int frame_scale_factor[FRAME_SCALE_STEPS] = {16, 24};
static const int frame_scale_factor[FRAME_SCALE_STEPS] = { 16, 24 };
// Multiplier of the target rate to be used as threshold for triggering scaling.
static const double rate_thresh_mult[FRAME_SCALE_STEPS] = {1.0, 2.0};
static const double rate_thresh_mult[FRAME_SCALE_STEPS] = { 1.0, 2.0 };
// Scale dependent Rate Correction Factor multipliers. Compensates for the
// greater number of bits per pixel generated in down-scaled frames.
static const double rcf_mult[FRAME_SCALE_STEPS] = {1.0, 2.0};
static const double rcf_mult[FRAME_SCALE_STEPS] = { 1.0, 2.0 };
typedef struct {
// Rate targetting variables
int base_frame_target; // A baseline frame target before adjustment
// for previous under or over shoot.
int this_frame_target; // Actual frame target after rc adjustment.
int base_frame_target; // A baseline frame target before adjustment
// for previous under or over shoot.
int this_frame_target; // Actual frame target after rc adjustment.
int projected_frame_size;
int sb64_target_rate;
int last_q[FRAME_TYPES]; // Separate values for Intra/Inter
int last_boosted_qindex; // Last boosted GF/KF/ARF q
int last_kf_qindex; // Q index of the last key frame coded.
int last_q[FRAME_TYPES]; // Separate values for Intra/Inter
int last_boosted_qindex; // Last boosted GF/KF/ARF q
int last_kf_qindex; // Q index of the last key frame coded.
int gfu_boost;
int last_boost;
@@ -179,8 +174,7 @@ void vp9_rc_init(const struct VP9EncoderConfig *oxcf, int pass,
RATE_CONTROL *rc);
int vp9_estimate_bits_at_q(FRAME_TYPE frame_kind, int q, int mbs,
double correction_factor,
vpx_bit_depth_t bit_depth);
double correction_factor, vpx_bit_depth_t bit_depth);
double vp9_convert_qindex_to_q(int qindex, vpx_bit_depth_t bit_depth);
@@ -241,8 +235,7 @@ void vp9_rc_compute_frame_size_bounds(const struct VP9_COMP *cpi,
int *frame_over_shoot_limit);
// Picks q and q bounds given the target for bits
int vp9_rc_pick_q_and_bounds(const struct VP9_COMP *cpi,
int *bottom_index,
int vp9_rc_pick_q_and_bounds(const struct VP9_COMP *cpi, int *bottom_index,
int *top_index);
// Estimates q to achieve a target bits per frame

200
vp9/encoder/vp9_rd.c
View File

@@ -40,7 +40,7 @@
#include "vp9/encoder/vp9_rd.h"
#include "vp9/encoder/vp9_tokenize.h"
#define RD_THRESH_POW 1.25
#define RD_THRESH_POW 1.25
// Factor to weigh the rate for switchable interp filters.
#define SWITCHABLE_INTERP_RATE_FACTOR 1
@@ -98,8 +98,7 @@ static void fill_token_costs(vp9_coeff_cost *c,
for (l = 0; l < BAND_COEFF_CONTEXTS(k); ++l) {
vpx_prob probs[ENTROPY_NODES];
vp9_model_to_full_probs(p[t][i][j][k][l], probs);
vp9_cost_tokens((int *)c[t][i][j][k][0][l], probs,
vp9_coef_tree);
vp9_cost_tokens((int *)c[t][i][j][k][0][l], probs, vp9_coef_tree);
vp9_cost_tokens_skip((int *)c[t][i][j][k][1][l], probs,
vp9_coef_tree);
assert(c[t][i][j][k][0][l][EOB_TOKEN] ==
@@ -142,28 +141,19 @@ void vp9_init_me_luts(void) {
#endif
}
static const int rd_boost_factor[16] = {
64, 32, 32, 32, 24, 16, 12, 12,
8, 8, 4, 4, 2, 2, 1, 0
};
static const int rd_frame_type_factor[FRAME_UPDATE_TYPES] = {
128, 144, 128, 128, 144
};
static const int rd_boost_factor[16] = { 64, 32, 32, 32, 24, 16, 12, 12,
8, 8, 4, 4, 2, 2, 1, 0 };
static const int rd_frame_type_factor[FRAME_UPDATE_TYPES] = { 128, 144, 128,
128, 144 };
int vp9_compute_rd_mult(const VP9_COMP *cpi, int qindex) {
const int64_t q = vp9_dc_quant(qindex, 0, cpi->common.bit_depth);
#if CONFIG_VP9_HIGHBITDEPTH
int64_t rdmult = 0;
switch (cpi->common.bit_depth) {
case VPX_BITS_8:
rdmult = 88 * q * q / 24;
break;
case VPX_BITS_10:
rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 4);
break;
case VPX_BITS_12:
rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 8);
break;
case VPX_BITS_8: rdmult = 88 * q * q / 24; break;
case VPX_BITS_10: rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 4); break;
case VPX_BITS_12: rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 8); break;
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
return -1;
@@ -179,8 +169,7 @@ int vp9_compute_rd_mult(const VP9_COMP *cpi, int qindex) {
rdmult = (rdmult * rd_frame_type_factor[frame_type]) >> 7;
rdmult += ((rdmult * rd_boost_factor[boost_index]) >> 7);
}
if (rdmult < 1)
rdmult = 1;
if (rdmult < 1) rdmult = 1;
return (int)rdmult;
}
@@ -188,21 +177,15 @@ static int compute_rd_thresh_factor(int qindex, vpx_bit_depth_t bit_depth) {
double q;
#if CONFIG_VP9_HIGHBITDEPTH
switch (bit_depth) {
case VPX_BITS_8:
q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
break;
case VPX_BITS_10:
q = vp9_dc_quant(qindex, 0, VPX_BITS_10) / 16.0;
break;
case VPX_BITS_12:
q = vp9_dc_quant(qindex, 0, VPX_BITS_12) / 64.0;
break;
case VPX_BITS_8: q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0; break;
case VPX_BITS_10: q = vp9_dc_quant(qindex, 0, VPX_BITS_10) / 16.0; break;
case VPX_BITS_12: q = vp9_dc_quant(qindex, 0, VPX_BITS_12) / 64.0; break;
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
return -1;
}
#else
(void) bit_depth;
(void)bit_depth;
q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
#endif // CONFIG_VP9_HIGHBITDEPTH
// TODO(debargha): Adjust the function below.
@@ -240,7 +223,8 @@ static void set_block_thresholds(const VP9_COMMON *cm, RD_OPT *rd) {
for (segment_id = 0; segment_id < MAX_SEGMENTS; ++segment_id) {
const int qindex =
clamp(vp9_get_qindex(&cm->seg, segment_id, cm->base_qindex) +
cm->y_dc_delta_q, 0, MAXQ);
cm->y_dc_delta_q,
0, MAXQ);
const int q = compute_rd_thresh_factor(qindex, cm->bit_depth);
for (bsize = 0; bsize < BLOCK_SIZES; ++bsize) {
@@ -251,10 +235,9 @@ static void set_block_thresholds(const VP9_COMMON *cm, RD_OPT *rd) {
if (bsize >= BLOCK_8X8) {
for (i = 0; i < MAX_MODES; ++i)
rd->threshes[segment_id][bsize][i] =
rd->thresh_mult[i] < thresh_max
? rd->thresh_mult[i] * t / 4
: INT_MAX;
rd->threshes[segment_id][bsize][i] = rd->thresh_mult[i] < thresh_max
? rd->thresh_mult[i] * t / 4
: INT_MAX;
} else {
for (i = 0; i < MAX_REFS; ++i)
rd->threshes[segment_id][bsize][i] =
@@ -281,7 +264,9 @@ void vp9_initialize_rd_consts(VP9_COMP *cpi) {
set_error_per_bit(x, rd->RDMULT);
x->select_tx_size = (cpi->sf.tx_size_search_method == USE_LARGESTALL &&
cm->frame_type != KEY_FRAME) ? 0 : 1;
cm->frame_type != KEY_FRAME)
? 0
: 1;
set_block_thresholds(cm, rd);
set_partition_probs(cm, xd);
@@ -335,19 +320,15 @@ static void model_rd_norm(int xsq_q10, int *r_q10, int *d_q10) {
// where r = exp(-sqrt(2) * x) and x = qpstep / sqrt(variance),
// and H(x) is the binary entropy function.
static const int rate_tab_q10[] = {
65536, 6086, 5574, 5275, 5063, 4899, 4764, 4651,
4553, 4389, 4255, 4142, 4044, 3958, 3881, 3811,
3748, 3635, 3538, 3453, 3376, 3307, 3244, 3186,
3133, 3037, 2952, 2877, 2809, 2747, 2690, 2638,
2589, 2501, 2423, 2353, 2290, 2232, 2179, 2130,
2084, 2001, 1928, 1862, 1802, 1748, 1698, 1651,
1608, 1530, 1460, 1398, 1342, 1290, 1243, 1199,
1159, 1086, 1021, 963, 911, 864, 821, 781,
745, 680, 623, 574, 530, 490, 455, 424,
395, 345, 304, 269, 239, 213, 190, 171,
154, 126, 104, 87, 73, 61, 52, 44,
38, 28, 21, 16, 12, 10, 8, 6,
5, 3, 2, 1, 1, 1, 0, 0,
65536, 6086, 5574, 5275, 5063, 4899, 4764, 4651, 4553, 4389, 4255, 4142,
4044, 3958, 3881, 3811, 3748, 3635, 3538, 3453, 3376, 3307, 3244, 3186,
3133, 3037, 2952, 2877, 2809, 2747, 2690, 2638, 2589, 2501, 2423, 2353,
2290, 2232, 2179, 2130, 2084, 2001, 1928, 1862, 1802, 1748, 1698, 1651,
1608, 1530, 1460, 1398, 1342, 1290, 1243, 1199, 1159, 1086, 1021, 963,
911, 864, 821, 781, 745, 680, 623, 574, 530, 490, 455, 424,
395, 345, 304, 269, 239, 213, 190, 171, 154, 126, 104, 87,
73, 61, 52, 44, 38, 28, 21, 16, 12, 10, 8, 6,
5, 3, 2, 1, 1, 1, 0, 0,
};
// Normalized distortion:
@@ -358,34 +339,29 @@ static void model_rd_norm(int xsq_q10, int *r_q10, int *d_q10) {
// where x = qpstep / sqrt(variance).
// Note the actual distortion is Dn * variance.
static const int dist_tab_q10[] = {
0, 0, 1, 1, 1, 2, 2, 2,
3, 3, 4, 5, 5, 6, 7, 7,
8, 9, 11, 12, 13, 15, 16, 17,
18, 21, 24, 26, 29, 31, 34, 36,
39, 44, 49, 54, 59, 64, 69, 73,
78, 88, 97, 106, 115, 124, 133, 142,
151, 167, 184, 200, 215, 231, 245, 260,
274, 301, 327, 351, 375, 397, 418, 439,
458, 495, 528, 559, 587, 613, 637, 659,
680, 717, 749, 777, 801, 823, 842, 859,
874, 899, 919, 936, 949, 960, 969, 977,
983, 994, 1001, 1006, 1010, 1013, 1015, 1017,
1018, 1020, 1022, 1022, 1023, 1023, 1023, 1024,
0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 5,
5, 6, 7, 7, 8, 9, 11, 12, 13, 15, 16, 17,
18, 21, 24, 26, 29, 31, 34, 36, 39, 44, 49, 54,
59, 64, 69, 73, 78, 88, 97, 106, 115, 124, 133, 142,
151, 167, 184, 200, 215, 231, 245, 260, 274, 301, 327, 351,
375, 397, 418, 439, 458, 495, 528, 559, 587, 613, 637, 659,
680, 717, 749, 777, 801, 823, 842, 859, 874, 899, 919, 936,
949, 960, 969, 977, 983, 994, 1001, 1006, 1010, 1013, 1015, 1017,
1018, 1020, 1022, 1022, 1023, 1023, 1023, 1024,
};
static const int xsq_iq_q10[] = {
0, 4, 8, 12, 16, 20, 24, 28,
32, 40, 48, 56, 64, 72, 80, 88,
96, 112, 128, 144, 160, 176, 192, 208,
224, 256, 288, 320, 352, 384, 416, 448,
480, 544, 608, 672, 736, 800, 864, 928,
992, 1120, 1248, 1376, 1504, 1632, 1760, 1888,
2016, 2272, 2528, 2784, 3040, 3296, 3552, 3808,
4064, 4576, 5088, 5600, 6112, 6624, 7136, 7648,
8160, 9184, 10208, 11232, 12256, 13280, 14304, 15328,
16352, 18400, 20448, 22496, 24544, 26592, 28640, 30688,
32736, 36832, 40928, 45024, 49120, 53216, 57312, 61408,
65504, 73696, 81888, 90080, 98272, 106464, 114656, 122848,
131040, 147424, 163808, 180192, 196576, 212960, 229344, 245728,
0, 4, 8, 12, 16, 20, 24, 28, 32,
40, 48, 56, 64, 72, 80, 88, 96, 112,
128, 144, 160, 176, 192, 208, 224, 256, 288,
320, 352, 384, 416, 448, 480, 544, 608, 672,
736, 800, 864, 928, 992, 1120, 1248, 1376, 1504,
1632, 1760, 1888, 2016, 2272, 2528, 2784, 3040, 3296,
3552, 3808, 4064, 4576, 5088, 5600, 6112, 6624, 7136,
7648, 8160, 9184, 10208, 11232, 12256, 13280, 14304, 15328,
16352, 18400, 20448, 22496, 24544, 26592, 28640, 30688, 32736,
36832, 40928, 45024, 49120, 53216, 57312, 61408, 65504, 73696,
81888, 90080, 98272, 106464, 114656, 122848, 131040, 147424, 163808,
180192, 196576, 212960, 229344, 245728,
};
const int tmp = (xsq_q10 >> 2) + 8;
const int k = get_msb(tmp) - 3;
@@ -455,15 +431,12 @@ void vp9_get_entropy_contexts(BLOCK_SIZE bsize, TX_SIZE tx_size,
for (i = 0; i < num_4x4_h; i += 8)
t_left[i] = !!*(const uint64_t *)&left[i];
break;
default:
assert(0 && "Invalid transform size.");
break;
default: assert(0 && "Invalid transform size."); break;
}
}
void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
uint8_t *ref_y_buffer, int ref_y_stride,
int ref_frame, BLOCK_SIZE block_size) {
void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x, uint8_t *ref_y_buffer,
int ref_y_stride, int ref_frame, BLOCK_SIZE block_size) {
int i;
int zero_seen = 0;
int best_index = 0;
@@ -473,9 +446,9 @@ void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
int near_same_nearest;
uint8_t *src_y_ptr = x->plane[0].src.buf;
uint8_t *ref_y_ptr;
const int num_mv_refs = MAX_MV_REF_CANDIDATES +
(cpi->sf.adaptive_motion_search &&
block_size < x->max_partition_size);
const int num_mv_refs =
MAX_MV_REF_CANDIDATES +
(cpi->sf.adaptive_motion_search && block_size < x->max_partition_size);
MV pred_mv[3];
pred_mv[0] = x->mbmi_ext->ref_mvs[ref_frame][0].as_mv;
@@ -483,25 +456,22 @@ void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
pred_mv[2] = x->pred_mv[ref_frame];
assert(num_mv_refs <= (int)(sizeof(pred_mv) / sizeof(pred_mv[0])));
near_same_nearest =
x->mbmi_ext->ref_mvs[ref_frame][0].as_int ==
x->mbmi_ext->ref_mvs[ref_frame][1].as_int;
near_same_nearest = x->mbmi_ext->ref_mvs[ref_frame][0].as_int ==
x->mbmi_ext->ref_mvs[ref_frame][1].as_int;
// Get the sad for each candidate reference mv.
for (i = 0; i < num_mv_refs; ++i) {
const MV *this_mv = &pred_mv[i];
int fp_row, fp_col;
if (i == 1 && near_same_nearest)
continue;
if (i == 1 && near_same_nearest) continue;
fp_row = (this_mv->row + 3 + (this_mv->row >= 0)) >> 3;
fp_col = (this_mv->col + 3 + (this_mv->col >= 0)) >> 3;
max_mv = VPXMAX(max_mv, VPXMAX(abs(this_mv->row), abs(this_mv->col)) >> 3);
if (fp_row ==0 && fp_col == 0 && zero_seen)
continue;
zero_seen |= (fp_row ==0 && fp_col == 0);
if (fp_row == 0 && fp_col == 0 && zero_seen) continue;
zero_seen |= (fp_row == 0 && fp_col == 0);
ref_y_ptr =&ref_y_buffer[ref_y_stride * fp_row + fp_col];
ref_y_ptr = &ref_y_buffer[ref_y_stride * fp_row + fp_col];
// Find sad for current vector.
this_sad = cpi->fn_ptr[block_size].sdf(src_y_ptr, x->plane[0].src.stride,
ref_y_ptr, ref_y_stride);
@@ -520,8 +490,7 @@ void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
void vp9_setup_pred_block(const MACROBLOCKD *xd,
struct buf_2d dst[MAX_MB_PLANE],
const YV12_BUFFER_CONFIG *src,
int mi_row, int mi_col,
const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
const struct scale_factors *scale,
const struct scale_factors *scale_uv) {
int i;
@@ -534,21 +503,21 @@ void vp9_setup_pred_block(const MACROBLOCKD *xd,
for (i = 0; i < MAX_MB_PLANE; ++i) {
setup_pred_plane(dst + i, dst[i].buf, dst[i].stride, mi_row, mi_col,
i ? scale_uv : scale,
xd->plane[i].subsampling_x, xd->plane[i].subsampling_y);
i ? scale_uv : scale, xd->plane[i].subsampling_x,
xd->plane[i].subsampling_y);
}
}
int vp9_raster_block_offset(BLOCK_SIZE plane_bsize,
int raster_block, int stride) {
int vp9_raster_block_offset(BLOCK_SIZE plane_bsize, int raster_block,
int stride) {
const int bw = b_width_log2_lookup[plane_bsize];
const int y = 4 * (raster_block >> bw);
const int x = 4 * (raster_block & ((1 << bw) - 1));
return y * stride + x;
}
int16_t* vp9_raster_block_offset_int16(BLOCK_SIZE plane_bsize,
int raster_block, int16_t *base) {
int16_t *vp9_raster_block_offset_int16(BLOCK_SIZE plane_bsize, int raster_block,
int16_t *base) {
const int stride = 4 * num_4x4_blocks_wide_lookup[plane_bsize];
return base + vp9_raster_block_offset(plane_bsize, raster_block, stride);
}
@@ -558,16 +527,16 @@ YV12_BUFFER_CONFIG *vp9_get_scaled_ref_frame(const VP9_COMP *cpi,
const VP9_COMMON *const cm = &cpi->common;
const int scaled_idx = cpi->scaled_ref_idx[ref_frame - 1];
const int ref_idx = get_ref_frame_buf_idx(cpi, ref_frame);
return
(scaled_idx != ref_idx && scaled_idx != INVALID_IDX) ?
&cm->buffer_pool->frame_bufs[scaled_idx].buf : NULL;
return (scaled_idx != ref_idx && scaled_idx != INVALID_IDX)
? &cm->buffer_pool->frame_bufs[scaled_idx].buf
: NULL;
}
int vp9_get_switchable_rate(const VP9_COMP *cpi, const MACROBLOCKD *const xd) {
const MODE_INFO *const mi = xd->mi[0];
const int ctx = get_pred_context_switchable_interp(xd);
return SWITCHABLE_INTERP_RATE_FACTOR *
cpi->switchable_interp_costs[ctx][mi->interp_filter];
cpi->switchable_interp_costs[ctx][mi->interp_filter];
}
void vp9_set_rd_speed_thresholds(VP9_COMP *cpi) {
@@ -616,7 +585,7 @@ void vp9_set_rd_speed_thresholds(VP9_COMP *cpi) {
rd->thresh_mult[THR_H_PRED] += 2000;
rd->thresh_mult[THR_V_PRED] += 2000;
rd->thresh_mult[THR_D45_PRED ] += 2500;
rd->thresh_mult[THR_D45_PRED] += 2500;
rd->thresh_mult[THR_D135_PRED] += 2500;
rd->thresh_mult[THR_D117_PRED] += 2500;
rd->thresh_mult[THR_D153_PRED] += 2500;
@@ -625,9 +594,10 @@ void vp9_set_rd_speed_thresholds(VP9_COMP *cpi) {
}
void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
static const int thresh_mult[2][MAX_REFS] =
{{2500, 2500, 2500, 4500, 4500, 2500},
{2000, 2000, 2000, 4000, 4000, 2000}};
static const int thresh_mult[2][MAX_REFS] = {
{ 2500, 2500, 2500, 4500, 4500, 2500 },
{ 2000, 2000, 2000, 4000, 4000, 2000 }
};
RD_OPT *const rd = &cpi->rd;
const int idx = cpi->oxcf.mode == BEST;
memcpy(rd->thresh_mult_sub8x8, thresh_mult[idx], sizeof(thresh_mult[idx]));
@@ -659,12 +629,9 @@ int vp9_get_intra_cost_penalty(int qindex, int qdelta,
const int q = vp9_dc_quant(qindex, qdelta, bit_depth);
#if CONFIG_VP9_HIGHBITDEPTH
switch (bit_depth) {
case VPX_BITS_8:
return 20 * q;
case VPX_BITS_10:
return 5 * q;
case VPX_BITS_12:
return ROUND_POWER_OF_TWO(5 * q, 2);
case VPX_BITS_8: return 20 * q;
case VPX_BITS_10: return 5 * q;
case VPX_BITS_12: return ROUND_POWER_OF_TWO(5 * q, 2);
default:
assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
return -1;
@@ -673,4 +640,3 @@ int vp9_get_intra_cost_penalty(int qindex, int qdelta,
return 20 * q;
#endif // CONFIG_VP9_HIGHBITDEPTH
}

39
vp9/encoder/vp9_rd.h
View File

@@ -23,23 +23,23 @@
extern "C" {
#endif
#define RDDIV_BITS 7
#define RD_EPB_SHIFT 6
#define RDDIV_BITS 7
#define RD_EPB_SHIFT 6
#define RDCOST(RM, DM, R, D) \
(ROUND_POWER_OF_TWO(((int64_t)R) * (RM), VP9_PROB_COST_SHIFT) + (D << DM))
#define QIDX_SKIP_THRESH 115
#define QIDX_SKIP_THRESH 115
#define MV_COST_WEIGHT 108
#define MV_COST_WEIGHT_SUB 120
#define MV_COST_WEIGHT 108
#define MV_COST_WEIGHT_SUB 120
#define INVALID_MV 0x80008000
#define MAX_MODES 30
#define MAX_REFS 6
#define MAX_REFS 6
#define RD_THRESH_MAX_FACT 64
#define RD_THRESH_INC 1
#define RD_THRESH_INC 1
// This enumerator type needs to be kept aligned with the mode order in
// const MODE_DEFINITION vp9_mode_order[MAX_MODES] used in the rd code.
@@ -135,17 +135,16 @@ void vp9_initialize_rd_consts(struct VP9_COMP *cpi);
void vp9_initialize_me_consts(struct VP9_COMP *cpi, MACROBLOCK *x, int qindex);
void vp9_model_rd_from_var_lapndz(unsigned int var, unsigned int n,
unsigned int qstep, int *rate,
int64_t *dist);
unsigned int qstep, int *rate, int64_t *dist);
int vp9_get_switchable_rate(const struct VP9_COMP *cpi,
const MACROBLOCKD *const xd);
int vp9_raster_block_offset(BLOCK_SIZE plane_bsize,
int raster_block, int stride);
int vp9_raster_block_offset(BLOCK_SIZE plane_bsize, int raster_block,
int stride);
int16_t* vp9_raster_block_offset_int16(BLOCK_SIZE plane_bsize,
int raster_block, int16_t *base);
int16_t *vp9_raster_block_offset_int16(BLOCK_SIZE plane_bsize, int raster_block,
int16_t *base);
YV12_BUFFER_CONFIG *vp9_get_scaled_ref_frame(const struct VP9_COMP *cpi,
int ref_frame);
@@ -161,12 +160,12 @@ void vp9_set_rd_speed_thresholds(struct VP9_COMP *cpi);
void vp9_set_rd_speed_thresholds_sub8x8(struct VP9_COMP *cpi);
void vp9_update_rd_thresh_fact(int (*fact)[MAX_MODES], int rd_thresh,
int bsize, int best_mode_index);
void vp9_update_rd_thresh_fact(int (*fact)[MAX_MODES], int rd_thresh, int bsize,
int best_mode_index);
static INLINE int rd_less_than_thresh(int64_t best_rd, int thresh,
int thresh_fact) {
return best_rd < ((int64_t)thresh * thresh_fact >> 5) || thresh == INT_MAX;
return best_rd < ((int64_t)thresh * thresh_fact >> 5) || thresh == INT_MAX;
}
static INLINE void set_error_per_bit(MACROBLOCK *x, int rdmult) {
@@ -174,14 +173,12 @@ static INLINE void set_error_per_bit(MACROBLOCK *x, int rdmult) {
x->errorperbit += (x->errorperbit == 0);
}
void vp9_mv_pred(struct VP9_COMP *cpi, MACROBLOCK *x,
uint8_t *ref_y_buffer, int ref_y_stride,
int ref_frame, BLOCK_SIZE block_size);
void vp9_mv_pred(struct VP9_COMP *cpi, MACROBLOCK *x, uint8_t *ref_y_buffer,
int ref_y_stride, int ref_frame, BLOCK_SIZE block_size);
void vp9_setup_pred_block(const MACROBLOCKD *xd,
struct buf_2d dst[MAX_MB_PLANE],
const YV12_BUFFER_CONFIG *src,
int mi_row, int mi_col,
const YV12_BUFFER_CONFIG *src, int mi_row, int mi_col,
const struct scale_factors *scale,
const struct scale_factors *scale_uv);

1568
vp9/encoder/vp9_rdopt.c
View File

File diff suppressed because it is too large Load Diff

26
vp9/encoder/vp9_rdopt.h
View File

@@ -31,19 +31,14 @@ void vp9_rd_pick_intra_mode_sb(struct VP9_COMP *cpi, struct macroblock *x,
void vp9_rd_pick_inter_mode_sb(struct VP9_COMP *cpi,
struct TileDataEnc *tile_data,
struct macroblock *x,
int mi_row, int mi_col,
struct RD_COST *rd_cost,
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx,
int64_t best_rd_so_far);
struct macroblock *x, int mi_row, int mi_col,
struct RD_COST *rd_cost, BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx, int64_t best_rd_so_far);
void vp9_rd_pick_inter_mode_sb_seg_skip(struct VP9_COMP *cpi,
struct TileDataEnc *tile_data,
struct macroblock *x,
struct RD_COST *rd_cost,
BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx,
int64_t best_rd_so_far);
void vp9_rd_pick_inter_mode_sb_seg_skip(
struct VP9_COMP *cpi, struct TileDataEnc *tile_data, struct macroblock *x,
struct RD_COST *rd_cost, BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx,
int64_t best_rd_so_far);
int vp9_internal_image_edge(struct VP9_COMP *cpi);
int vp9_active_h_edge(struct VP9_COMP *cpi, int mi_row, int mi_step);
@@ -52,10 +47,9 @@ int vp9_active_edge_sb(struct VP9_COMP *cpi, int mi_row, int mi_col);
void vp9_rd_pick_inter_mode_sub8x8(struct VP9_COMP *cpi,
struct TileDataEnc *tile_data,
struct macroblock *x,
int mi_row, int mi_col,
struct RD_COST *rd_cost,
BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx,
struct macroblock *x, int mi_row, int mi_col,
struct RD_COST *rd_cost, BLOCK_SIZE bsize,
PICK_MODE_CONTEXT *ctx,
int64_t best_rd_so_far);
#ifdef __cplusplus

523
vp9/encoder/vp9_resize.c
View File

@@ -23,198 +23,118 @@
#include "vp9/common/vp9_common.h"
#include "vp9/encoder/vp9_resize.h"
#define FILTER_BITS 7
#define FILTER_BITS 7
#define INTERP_TAPS 8
#define SUBPEL_BITS 5
#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
#define INTERP_PRECISION_BITS 32
#define INTERP_TAPS 8
#define SUBPEL_BITS 5
#define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
#define INTERP_PRECISION_BITS 32
typedef int16_t interp_kernel[INTERP_TAPS];
// Filters for interpolation (0.5-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters500[(1 << SUBPEL_BITS)] = {
{-3, 0, 35, 64, 35, 0, -3, 0},
{-3, -1, 34, 64, 36, 1, -3, 0},
{-3, -1, 32, 64, 38, 1, -3, 0},
{-2, -2, 31, 63, 39, 2, -3, 0},
{-2, -2, 29, 63, 41, 2, -3, 0},
{-2, -2, 28, 63, 42, 3, -4, 0},
{-2, -3, 27, 63, 43, 4, -4, 0},
{-2, -3, 25, 62, 45, 5, -4, 0},
{-2, -3, 24, 62, 46, 5, -4, 0},
{-2, -3, 23, 61, 47, 6, -4, 0},
{-2, -3, 21, 60, 49, 7, -4, 0},
{-1, -4, 20, 60, 50, 8, -4, -1},
{-1, -4, 19, 59, 51, 9, -4, -1},
{-1, -4, 17, 58, 52, 10, -4, 0},
{-1, -4, 16, 57, 53, 12, -4, -1},
{-1, -4, 15, 56, 54, 13, -4, -1},
{-1, -4, 14, 55, 55, 14, -4, -1},
{-1, -4, 13, 54, 56, 15, -4, -1},
{-1, -4, 12, 53, 57, 16, -4, -1},
{0, -4, 10, 52, 58, 17, -4, -1},
{-1, -4, 9, 51, 59, 19, -4, -1},
{-1, -4, 8, 50, 60, 20, -4, -1},
{0, -4, 7, 49, 60, 21, -3, -2},
{0, -4, 6, 47, 61, 23, -3, -2},
{0, -4, 5, 46, 62, 24, -3, -2},
{0, -4, 5, 45, 62, 25, -3, -2},
{0, -4, 4, 43, 63, 27, -3, -2},
{0, -4, 3, 42, 63, 28, -2, -2},
{0, -3, 2, 41, 63, 29, -2, -2},
{0, -3, 2, 39, 63, 31, -2, -2},
{0, -3, 1, 38, 64, 32, -1, -3},
{0, -3, 1, 36, 64, 34, -1, -3}
{ -3, 0, 35, 64, 35, 0, -3, 0 }, { -3, -1, 34, 64, 36, 1, -3, 0 },
{ -3, -1, 32, 64, 38, 1, -3, 0 }, { -2, -2, 31, 63, 39, 2, -3, 0 },
{ -2, -2, 29, 63, 41, 2, -3, 0 }, { -2, -2, 28, 63, 42, 3, -4, 0 },
{ -2, -3, 27, 63, 43, 4, -4, 0 }, { -2, -3, 25, 62, 45, 5, -4, 0 },
{ -2, -3, 24, 62, 46, 5, -4, 0 }, { -2, -3, 23, 61, 47, 6, -4, 0 },
{ -2, -3, 21, 60, 49, 7, -4, 0 }, { -1, -4, 20, 60, 50, 8, -4, -1 },
{ -1, -4, 19, 59, 51, 9, -4, -1 }, { -1, -4, 17, 58, 52, 10, -4, 0 },
{ -1, -4, 16, 57, 53, 12, -4, -1 }, { -1, -4, 15, 56, 54, 13, -4, -1 },
{ -1, -4, 14, 55, 55, 14, -4, -1 }, { -1, -4, 13, 54, 56, 15, -4, -1 },
{ -1, -4, 12, 53, 57, 16, -4, -1 }, { 0, -4, 10, 52, 58, 17, -4, -1 },
{ -1, -4, 9, 51, 59, 19, -4, -1 }, { -1, -4, 8, 50, 60, 20, -4, -1 },
{ 0, -4, 7, 49, 60, 21, -3, -2 }, { 0, -4, 6, 47, 61, 23, -3, -2 },
{ 0, -4, 5, 46, 62, 24, -3, -2 }, { 0, -4, 5, 45, 62, 25, -3, -2 },
{ 0, -4, 4, 43, 63, 27, -3, -2 }, { 0, -4, 3, 42, 63, 28, -2, -2 },
{ 0, -3, 2, 41, 63, 29, -2, -2 }, { 0, -3, 2, 39, 63, 31, -2, -2 },
{ 0, -3, 1, 38, 64, 32, -1, -3 }, { 0, -3, 1, 36, 64, 34, -1, -3 }
};
// Filters for interpolation (0.625-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters625[(1 << SUBPEL_BITS)] = {
{-1, -8, 33, 80, 33, -8, -1, 0},
{-1, -8, 30, 80, 35, -8, -1, 1},
{-1, -8, 28, 80, 37, -7, -2, 1},
{0, -8, 26, 79, 39, -7, -2, 1},
{0, -8, 24, 79, 41, -7, -2, 1},
{0, -8, 22, 78, 43, -6, -2, 1},
{0, -8, 20, 78, 45, -5, -3, 1},
{0, -8, 18, 77, 48, -5, -3, 1},
{0, -8, 16, 76, 50, -4, -3, 1},
{0, -8, 15, 75, 52, -3, -4, 1},
{0, -7, 13, 74, 54, -3, -4, 1},
{0, -7, 11, 73, 56, -2, -4, 1},
{0, -7, 10, 71, 58, -1, -4, 1},
{1, -7, 8, 70, 60, 0, -5, 1},
{1, -6, 6, 68, 62, 1, -5, 1},
{1, -6, 5, 67, 63, 2, -5, 1},
{1, -6, 4, 65, 65, 4, -6, 1},
{1, -5, 2, 63, 67, 5, -6, 1},
{1, -5, 1, 62, 68, 6, -6, 1},
{1, -5, 0, 60, 70, 8, -7, 1},
{1, -4, -1, 58, 71, 10, -7, 0},
{1, -4, -2, 56, 73, 11, -7, 0},
{1, -4, -3, 54, 74, 13, -7, 0},
{1, -4, -3, 52, 75, 15, -8, 0},
{1, -3, -4, 50, 76, 16, -8, 0},
{1, -3, -5, 48, 77, 18, -8, 0},
{1, -3, -5, 45, 78, 20, -8, 0},
{1, -2, -6, 43, 78, 22, -8, 0},
{1, -2, -7, 41, 79, 24, -8, 0},
{1, -2, -7, 39, 79, 26, -8, 0},
{1, -2, -7, 37, 80, 28, -8, -1},
{1, -1, -8, 35, 80, 30, -8, -1},
{ -1, -8, 33, 80, 33, -8, -1, 0 }, { -1, -8, 30, 80, 35, -8, -1, 1 },
{ -1, -8, 28, 80, 37, -7, -2, 1 }, { 0, -8, 26, 79, 39, -7, -2, 1 },
{ 0, -8, 24, 79, 41, -7, -2, 1 }, { 0, -8, 22, 78, 43, -6, -2, 1 },
{ 0, -8, 20, 78, 45, -5, -3, 1 }, { 0, -8, 18, 77, 48, -5, -3, 1 },
{ 0, -8, 16, 76, 50, -4, -3, 1 }, { 0, -8, 15, 75, 52, -3, -4, 1 },
{ 0, -7, 13, 74, 54, -3, -4, 1 }, { 0, -7, 11, 73, 56, -2, -4, 1 },
{ 0, -7, 10, 71, 58, -1, -4, 1 }, { 1, -7, 8, 70, 60, 0, -5, 1 },
{ 1, -6, 6, 68, 62, 1, -5, 1 }, { 1, -6, 5, 67, 63, 2, -5, 1 },
{ 1, -6, 4, 65, 65, 4, -6, 1 }, { 1, -5, 2, 63, 67, 5, -6, 1 },
{ 1, -5, 1, 62, 68, 6, -6, 1 }, { 1, -5, 0, 60, 70, 8, -7, 1 },
{ 1, -4, -1, 58, 71, 10, -7, 0 }, { 1, -4, -2, 56, 73, 11, -7, 0 },
{ 1, -4, -3, 54, 74, 13, -7, 0 }, { 1, -4, -3, 52, 75, 15, -8, 0 },
{ 1, -3, -4, 50, 76, 16, -8, 0 }, { 1, -3, -5, 48, 77, 18, -8, 0 },
{ 1, -3, -5, 45, 78, 20, -8, 0 }, { 1, -2, -6, 43, 78, 22, -8, 0 },
{ 1, -2, -7, 41, 79, 24, -8, 0 }, { 1, -2, -7, 39, 79, 26, -8, 0 },
{ 1, -2, -7, 37, 80, 28, -8, -1 }, { 1, -1, -8, 35, 80, 30, -8, -1 },
};
// Filters for interpolation (0.75-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters750[(1 << SUBPEL_BITS)] = {
{2, -11, 25, 96, 25, -11, 2, 0},
{2, -11, 22, 96, 28, -11, 2, 0},
{2, -10, 19, 95, 31, -11, 2, 0},
{2, -10, 17, 95, 34, -12, 2, 0},
{2, -9, 14, 94, 37, -12, 2, 0},
{2, -8, 12, 93, 40, -12, 1, 0},
{2, -8, 9, 92, 43, -12, 1, 1},
{2, -7, 7, 91, 46, -12, 1, 0},
{2, -7, 5, 90, 49, -12, 1, 0},
{2, -6, 3, 88, 52, -12, 0, 1},
{2, -5, 1, 86, 55, -12, 0, 1},
{2, -5, -1, 84, 58, -11, 0, 1},
{2, -4, -2, 82, 61, -11, -1, 1},
{2, -4, -4, 80, 64, -10, -1, 1},
{1, -3, -5, 77, 67, -9, -1, 1},
{1, -3, -6, 75, 70, -8, -2, 1},
{1, -2, -7, 72, 72, -7, -2, 1},
{1, -2, -8, 70, 75, -6, -3, 1},
{1, -1, -9, 67, 77, -5, -3, 1},
{1, -1, -10, 64, 80, -4, -4, 2},
{1, -1, -11, 61, 82, -2, -4, 2},
{1, 0, -11, 58, 84, -1, -5, 2},
{1, 0, -12, 55, 86, 1, -5, 2},
{1, 0, -12, 52, 88, 3, -6, 2},
{0, 1, -12, 49, 90, 5, -7, 2},
{0, 1, -12, 46, 91, 7, -7, 2},
{1, 1, -12, 43, 92, 9, -8, 2},
{0, 1, -12, 40, 93, 12, -8, 2},
{0, 2, -12, 37, 94, 14, -9, 2},
{0, 2, -12, 34, 95, 17, -10, 2},
{0, 2, -11, 31, 95, 19, -10, 2},
{0, 2, -11, 28, 96, 22, -11, 2}
{ 2, -11, 25, 96, 25, -11, 2, 0 }, { 2, -11, 22, 96, 28, -11, 2, 0 },
{ 2, -10, 19, 95, 31, -11, 2, 0 }, { 2, -10, 17, 95, 34, -12, 2, 0 },
{ 2, -9, 14, 94, 37, -12, 2, 0 }, { 2, -8, 12, 93, 40, -12, 1, 0 },
{ 2, -8, 9, 92, 43, -12, 1, 1 }, { 2, -7, 7, 91, 46, -12, 1, 0 },
{ 2, -7, 5, 90, 49, -12, 1, 0 }, { 2, -6, 3, 88, 52, -12, 0, 1 },
{ 2, -5, 1, 86, 55, -12, 0, 1 }, { 2, -5, -1, 84, 58, -11, 0, 1 },
{ 2, -4, -2, 82, 61, -11, -1, 1 }, { 2, -4, -4, 80, 64, -10, -1, 1 },
{ 1, -3, -5, 77, 67, -9, -1, 1 }, { 1, -3, -6, 75, 70, -8, -2, 1 },
{ 1, -2, -7, 72, 72, -7, -2, 1 }, { 1, -2, -8, 70, 75, -6, -3, 1 },
{ 1, -1, -9, 67, 77, -5, -3, 1 }, { 1, -1, -10, 64, 80, -4, -4, 2 },
{ 1, -1, -11, 61, 82, -2, -4, 2 }, { 1, 0, -11, 58, 84, -1, -5, 2 },
{ 1, 0, -12, 55, 86, 1, -5, 2 }, { 1, 0, -12, 52, 88, 3, -6, 2 },
{ 0, 1, -12, 49, 90, 5, -7, 2 }, { 0, 1, -12, 46, 91, 7, -7, 2 },
{ 1, 1, -12, 43, 92, 9, -8, 2 }, { 0, 1, -12, 40, 93, 12, -8, 2 },
{ 0, 2, -12, 37, 94, 14, -9, 2 }, { 0, 2, -12, 34, 95, 17, -10, 2 },
{ 0, 2, -11, 31, 95, 19, -10, 2 }, { 0, 2, -11, 28, 96, 22, -11, 2 }
};
// Filters for interpolation (0.875-band) - note this also filters integer pels.
static const interp_kernel filteredinterp_filters875[(1 << SUBPEL_BITS)] = {
{3, -8, 13, 112, 13, -8, 3, 0},
{3, -7, 10, 112, 17, -9, 3, -1},
{2, -6, 7, 111, 21, -9, 3, -1},
{2, -5, 4, 111, 24, -10, 3, -1},
{2, -4, 1, 110, 28, -11, 3, -1},
{1, -3, -1, 108, 32, -12, 4, -1},
{1, -2, -3, 106, 36, -13, 4, -1},
{1, -1, -6, 105, 40, -14, 4, -1},
{1, -1, -7, 102, 44, -14, 4, -1},
{1, 0, -9, 100, 48, -15, 4, -1},
{1, 1, -11, 97, 53, -16, 4, -1},
{0, 1, -12, 95, 57, -16, 4, -1},
{0, 2, -13, 91, 61, -16, 4, -1},
{0, 2, -14, 88, 65, -16, 4, -1},
{0, 3, -15, 84, 69, -17, 4, 0},
{0, 3, -16, 81, 73, -16, 3, 0},
{0, 3, -16, 77, 77, -16, 3, 0},
{0, 3, -16, 73, 81, -16, 3, 0},
{0, 4, -17, 69, 84, -15, 3, 0},
{-1, 4, -16, 65, 88, -14, 2, 0},
{-1, 4, -16, 61, 91, -13, 2, 0},
{-1, 4, -16, 57, 95, -12, 1, 0},
{-1, 4, -16, 53, 97, -11, 1, 1},
{-1, 4, -15, 48, 100, -9, 0, 1},
{-1, 4, -14, 44, 102, -7, -1, 1},
{-1, 4, -14, 40, 105, -6, -1, 1},
{-1, 4, -13, 36, 106, -3, -2, 1},
{-1, 4, -12, 32, 108, -1, -3, 1},
{-1, 3, -11, 28, 110, 1, -4, 2},
{-1, 3, -10, 24, 111, 4, -5, 2},
{-1, 3, -9, 21, 111, 7, -6, 2},
{-1, 3, -9, 17, 112, 10, -7, 3}
{ 3, -8, 13, 112, 13, -8, 3, 0 }, { 3, -7, 10, 112, 17, -9, 3, -1 },
{ 2, -6, 7, 111, 21, -9, 3, -1 }, { 2, -5, 4, 111, 24, -10, 3, -1 },
{ 2, -4, 1, 110, 28, -11, 3, -1 }, { 1, -3, -1, 108, 32, -12, 4, -1 },
{ 1, -2, -3, 106, 36, -13, 4, -1 }, { 1, -1, -6, 105, 40, -14, 4, -1 },
{ 1, -1, -7, 102, 44, -14, 4, -1 }, { 1, 0, -9, 100, 48, -15, 4, -1 },
{ 1, 1, -11, 97, 53, -16, 4, -1 }, { 0, 1, -12, 95, 57, -16, 4, -1 },
{ 0, 2, -13, 91, 61, -16, 4, -1 }, { 0, 2, -14, 88, 65, -16, 4, -1 },
{ 0, 3, -15, 84, 69, -17, 4, 0 }, { 0, 3, -16, 81, 73, -16, 3, 0 },
{ 0, 3, -16, 77, 77, -16, 3, 0 }, { 0, 3, -16, 73, 81, -16, 3, 0 },
{ 0, 4, -17, 69, 84, -15, 3, 0 }, { -1, 4, -16, 65, 88, -14, 2, 0 },
{ -1, 4, -16, 61, 91, -13, 2, 0 }, { -1, 4, -16, 57, 95, -12, 1, 0 },
{ -1, 4, -16, 53, 97, -11, 1, 1 }, { -1, 4, -15, 48, 100, -9, 0, 1 },
{ -1, 4, -14, 44, 102, -7, -1, 1 }, { -1, 4, -14, 40, 105, -6, -1, 1 },
{ -1, 4, -13, 36, 106, -3, -2, 1 }, { -1, 4, -12, 32, 108, -1, -3, 1 },
{ -1, 3, -11, 28, 110, 1, -4, 2 }, { -1, 3, -10, 24, 111, 4, -5, 2 },
{ -1, 3, -9, 21, 111, 7, -6, 2 }, { -1, 3, -9, 17, 112, 10, -7, 3 }
};
// Filters for interpolation (full-band) - no filtering for integer pixels
static const interp_kernel filteredinterp_filters1000[(1 << SUBPEL_BITS)] = {
{0, 0, 0, 128, 0, 0, 0, 0},
{0, 1, -3, 128, 3, -1, 0, 0},
{-1, 2, -6, 127, 7, -2, 1, 0},
{-1, 3, -9, 126, 12, -4, 1, 0},
{-1, 4, -12, 125, 16, -5, 1, 0},
{-1, 4, -14, 123, 20, -6, 2, 0},
{-1, 5, -15, 120, 25, -8, 2, 0},
{-1, 5, -17, 118, 30, -9, 3, -1},
{-1, 6, -18, 114, 35, -10, 3, -1},
{-1, 6, -19, 111, 41, -12, 3, -1},
{-1, 6, -20, 107, 46, -13, 4, -1},
{-1, 6, -21, 103, 52, -14, 4, -1},
{-1, 6, -21, 99, 57, -16, 5, -1},
{-1, 6, -21, 94, 63, -17, 5, -1},
{-1, 6, -20, 89, 68, -18, 5, -1},
{-1, 6, -20, 84, 73, -19, 6, -1},
{-1, 6, -20, 79, 79, -20, 6, -1},
{-1, 6, -19, 73, 84, -20, 6, -1},
{-1, 5, -18, 68, 89, -20, 6, -1},
{-1, 5, -17, 63, 94, -21, 6, -1},
{-1, 5, -16, 57, 99, -21, 6, -1},
{-1, 4, -14, 52, 103, -21, 6, -1},
{-1, 4, -13, 46, 107, -20, 6, -1},
{-1, 3, -12, 41, 111, -19, 6, -1},
{-1, 3, -10, 35, 114, -18, 6, -1},
{-1, 3, -9, 30, 118, -17, 5, -1},
{0, 2, -8, 25, 120, -15, 5, -1},
{0, 2, -6, 20, 123, -14, 4, -1},
{0, 1, -5, 16, 125, -12, 4, -1},
{0, 1, -4, 12, 126, -9, 3, -1},
{0, 1, -2, 7, 127, -6, 2, -1},
{0, 0, -1, 3, 128, -3, 1, 0}
{ 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -3, 128, 3, -1, 0, 0 },
{ -1, 2, -6, 127, 7, -2, 1, 0 }, { -1, 3, -9, 126, 12, -4, 1, 0 },
{ -1, 4, -12, 125, 16, -5, 1, 0 }, { -1, 4, -14, 123, 20, -6, 2, 0 },
{ -1, 5, -15, 120, 25, -8, 2, 0 }, { -1, 5, -17, 118, 30, -9, 3, -1 },
{ -1, 6, -18, 114, 35, -10, 3, -1 }, { -1, 6, -19, 111, 41, -12, 3, -1 },
{ -1, 6, -20, 107, 46, -13, 4, -1 }, { -1, 6, -21, 103, 52, -14, 4, -1 },
{ -1, 6, -21, 99, 57, -16, 5, -1 }, { -1, 6, -21, 94, 63, -17, 5, -1 },
{ -1, 6, -20, 89, 68, -18, 5, -1 }, { -1, 6, -20, 84, 73, -19, 6, -1 },
{ -1, 6, -20, 79, 79, -20, 6, -1 }, { -1, 6, -19, 73, 84, -20, 6, -1 },
{ -1, 5, -18, 68, 89, -20, 6, -1 }, { -1, 5, -17, 63, 94, -21, 6, -1 },
{ -1, 5, -16, 57, 99, -21, 6, -1 }, { -1, 4, -14, 52, 103, -21, 6, -1 },
{ -1, 4, -13, 46, 107, -20, 6, -1 }, { -1, 3, -12, 41, 111, -19, 6, -1 },
{ -1, 3, -10, 35, 114, -18, 6, -1 }, { -1, 3, -9, 30, 118, -17, 5, -1 },
{ 0, 2, -8, 25, 120, -15, 5, -1 }, { 0, 2, -6, 20, 123, -14, 4, -1 },
{ 0, 1, -5, 16, 125, -12, 4, -1 }, { 0, 1, -4, 12, 126, -9, 3, -1 },
{ 0, 1, -2, 7, 127, -6, 2, -1 }, { 0, 0, -1, 3, 128, -3, 1, 0 }
};
// Filters for factor of 2 downsampling.
static const int16_t vp9_down2_symeven_half_filter[] = {56, 12, -3, -1};
static const int16_t vp9_down2_symodd_half_filter[] = {64, 35, 0, -3};
static const int16_t vp9_down2_symeven_half_filter[] = { 56, 12, -3, -1 };
static const int16_t vp9_down2_symodd_half_filter[] = { 64, 35, 0, -3 };
static const interp_kernel *choose_interp_filter(int inlength, int outlength) {
int outlength16 = outlength * 16;
@@ -232,11 +152,14 @@ static const interp_kernel *choose_interp_filter(int inlength, int outlength) {
static void interpolate(const uint8_t *const input, int inlength,
uint8_t *output, int outlength) {
const int64_t delta = (((uint64_t)inlength << 32) + outlength / 2) /
outlength;
const int64_t offset = inlength > outlength ?
(((int64_t)(inlength - outlength) << 31) + outlength / 2) / outlength :
-(((int64_t)(outlength - inlength) << 31) + outlength / 2) / outlength;
const int64_t delta =
(((uint64_t)inlength << 32) + outlength / 2) / outlength;
const int64_t offset =
inlength > outlength
? (((int64_t)(inlength - outlength) << 31) + outlength / 2) /
outlength
: -(((int64_t)(outlength - inlength) << 31) + outlength / 2) /
outlength;
uint8_t *optr = output;
int x, x1, x2, sum, k, int_pel, sub_pel;
int64_t y;
@@ -253,8 +176,8 @@ static void interpolate(const uint8_t *const input, int inlength,
x1 = x;
x = outlength - 1;
y = delta * x + offset;
while ((y >> INTERP_PRECISION_BITS) +
(int64_t)(INTERP_TAPS / 2) >= inlength) {
while ((y >> INTERP_PRECISION_BITS) + (int64_t)(INTERP_TAPS / 2) >=
inlength) {
x--;
y -= delta;
}
@@ -268,8 +191,8 @@ static void interpolate(const uint8_t *const input, int inlength,
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k) {
const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
sum += filter[k] * input[(pk < 0 ? 0 :
(pk >= inlength ? inlength - 1 : pk))];
sum += filter[k] *
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
}
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
@@ -282,9 +205,9 @@ static void interpolate(const uint8_t *const input, int inlength,
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0 ?
0 :
int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
? 0
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
// Middle part.
@@ -306,9 +229,9 @@ static void interpolate(const uint8_t *const input, int inlength,
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >=
inlength ? inlength - 1 :
int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
? inlength - 1
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
}
}
@@ -332,7 +255,7 @@ static void down2_symeven(const uint8_t *const input, int length,
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
@@ -362,7 +285,7 @@ static void down2_symeven(const uint8_t *const input, int length,
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
@@ -388,7 +311,7 @@ static void down2_symodd(const uint8_t *const input, int length,
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
@@ -417,7 +340,7 @@ static void down2_symodd(const uint8_t *const input, int length,
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel(sum);
@@ -427,8 +350,7 @@ static void down2_symodd(const uint8_t *const input, int length,
static int get_down2_length(int length, int steps) {
int s;
for (s = 0; s < steps; ++s)
length = (length + 1) >> 1;
for (s = 0; s < steps; ++s) length = (length + 1) >> 1;
return length;
}
@@ -442,11 +364,8 @@ static int get_down2_steps(int in_length, int out_length) {
return steps;
}
static void resize_multistep(const uint8_t *const input,
int length,
uint8_t *output,
int olength,
uint8_t *otmp) {
static void resize_multistep(const uint8_t *const input, int length,
uint8_t *output, int olength, uint8_t *otmp) {
int steps;
if (length == olength) {
memcpy(output, input, sizeof(output[0]) * length);
@@ -501,37 +420,31 @@ static void fill_arr_to_col(uint8_t *img, int stride, int len, uint8_t *arr) {
}
}
void vp9_resize_plane(const uint8_t *const input,
int height,
int width,
int in_stride,
uint8_t *output,
int height2,
int width2,
void vp9_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2, int width2,
int out_stride) {
int i;
uint8_t *intbuf = (uint8_t *)malloc(sizeof(uint8_t) * width2 * height);
uint8_t *tmpbuf = (uint8_t *)malloc(sizeof(uint8_t) *
(width < height ? height : width));
uint8_t *tmpbuf =
(uint8_t *)malloc(sizeof(uint8_t) * (width < height ? height : width));
uint8_t *arrbuf = (uint8_t *)malloc(sizeof(uint8_t) * height);
uint8_t *arrbuf2 = (uint8_t *)malloc(sizeof(uint8_t) * height2);
if (intbuf == NULL || tmpbuf == NULL ||
arrbuf == NULL || arrbuf2 == NULL)
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
goto Error;
assert(width > 0);
assert(height > 0);
assert(width2 > 0);
assert(height2 > 0);
for (i = 0; i < height; ++i)
resize_multistep(input + in_stride * i, width,
intbuf + width2 * i, width2, tmpbuf);
resize_multistep(input + in_stride * i, width, intbuf + width2 * i, width2,
tmpbuf);
for (i = 0; i < width2; ++i) {
fill_col_to_arr(intbuf + i, width2, height, arrbuf);
resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf);
fill_arr_to_col(output + i, out_stride, height2, arrbuf2);
}
Error:
Error:
free(intbuf);
free(tmpbuf);
free(arrbuf);
@@ -543,9 +456,12 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
uint16_t *output, int outlength, int bd) {
const int64_t delta =
(((uint64_t)inlength << 32) + outlength / 2) / outlength;
const int64_t offset = inlength > outlength ?
(((int64_t)(inlength - outlength) << 31) + outlength / 2) / outlength :
-(((int64_t)(outlength - inlength) << 31) + outlength / 2) / outlength;
const int64_t offset =
inlength > outlength
? (((int64_t)(inlength - outlength) << 31) + outlength / 2) /
outlength
: -(((int64_t)(outlength - inlength) << 31) + outlength / 2) /
outlength;
uint16_t *optr = output;
int x, x1, x2, sum, k, int_pel, sub_pel;
int64_t y;
@@ -562,8 +478,8 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
x1 = x;
x = outlength - 1;
y = delta * x + offset;
while ((y >> INTERP_PRECISION_BITS) +
(int64_t)(INTERP_TAPS / 2) >= inlength) {
while ((y >> INTERP_PRECISION_BITS) + (int64_t)(INTERP_TAPS / 2) >=
inlength) {
x--;
y -= delta;
}
@@ -578,7 +494,7 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
for (k = 0; k < INTERP_TAPS; ++k) {
const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
sum += filter[k] *
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
}
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
@@ -591,9 +507,9 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] *
input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0 ?
0 : int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
? 0
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
// Middle part.
@@ -615,9 +531,9 @@ static void highbd_interpolate(const uint16_t *const input, int inlength,
filter = interp_filters[sub_pel];
sum = 0;
for (k = 0; k < INTERP_TAPS; ++k)
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >=
inlength ? inlength - 1 :
int_pel - INTERP_TAPS / 2 + 1 + k)];
sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
? inlength - 1
: int_pel - INTERP_TAPS / 2 + 1 + k)];
*optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
}
}
@@ -641,7 +557,7 @@ static void highbd_down2_symeven(const uint16_t *const input, int length,
for (j = 0; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@@ -671,7 +587,7 @@ static void highbd_down2_symeven(const uint16_t *const input, int length,
for (j = 0; j < filter_len_half; ++j) {
sum += (input[i - j] +
input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@@ -680,7 +596,7 @@ static void highbd_down2_symeven(const uint16_t *const input, int length,
}
static void highbd_down2_symodd(const uint16_t *const input, int length,
uint16_t *output, int bd) {
uint16_t *output, int bd) {
// Actual filter len = 2 * filter_len_half - 1.
static const int16_t *filter = vp9_down2_symodd_half_filter;
const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
@@ -697,7 +613,7 @@ static void highbd_down2_symodd(const uint16_t *const input, int length,
for (j = 1; j < filter_len_half; ++j) {
sum += (input[(i - j < 0 ? 0 : i - j)] +
input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@@ -726,7 +642,7 @@ static void highbd_down2_symodd(const uint16_t *const input, int length,
int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
for (j = 1; j < filter_len_half; ++j) {
sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
filter[j];
filter[j];
}
sum >>= FILTER_BITS;
*optr++ = clip_pixel_highbd(sum, bd);
@@ -734,12 +650,9 @@ static void highbd_down2_symodd(const uint16_t *const input, int length,
}
}
static void highbd_resize_multistep(const uint16_t *const input,
int length,
uint16_t *output,
int olength,
uint16_t *otmp,
int bd) {
static void highbd_resize_multistep(const uint16_t *const input, int length,
uint16_t *output, int olength,
uint16_t *otmp, int bd) {
int steps;
if (length == olength) {
memcpy(output, input, sizeof(output[0]) * length);
@@ -796,23 +709,16 @@ static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len,
}
}
void vp9_highbd_resize_plane(const uint8_t *const input,
int height,
int width,
int in_stride,
uint8_t *output,
int height2,
int width2,
int out_stride,
int bd) {
void vp9_highbd_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2,
int width2, int out_stride, int bd) {
int i;
uint16_t *intbuf = (uint16_t *)malloc(sizeof(uint16_t) * width2 * height);
uint16_t *tmpbuf = (uint16_t *)malloc(sizeof(uint16_t) *
(width < height ? height : width));
uint16_t *tmpbuf =
(uint16_t *)malloc(sizeof(uint16_t) * (width < height ? height : width));
uint16_t *arrbuf = (uint16_t *)malloc(sizeof(uint16_t) * height);
uint16_t *arrbuf2 = (uint16_t *)malloc(sizeof(uint16_t) * height2);
if (intbuf == NULL || tmpbuf == NULL ||
arrbuf == NULL || arrbuf2 == NULL)
if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
goto Error;
for (i = 0; i < height; ++i) {
highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width,
@@ -820,13 +726,12 @@ void vp9_highbd_resize_plane(const uint8_t *const input,
}
for (i = 0; i < width2; ++i) {
highbd_fill_col_to_arr(intbuf + i, width2, height, arrbuf);
highbd_resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf,
bd);
highbd_resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf, bd);
highbd_fill_arr_to_col(CONVERT_TO_SHORTPTR(output + i), out_stride, height2,
arrbuf2);
}
Error:
Error:
free(intbuf);
free(tmpbuf);
free(arrbuf);
@@ -834,96 +739,82 @@ void vp9_highbd_resize_plane(const uint8_t *const input,
}
#endif // CONFIG_VP9_HIGHBITDEPTH
void vp9_resize_frame420(const uint8_t *const y,
int y_stride,
void vp9_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride,
int height, int width,
uint8_t *oy, int oy_stride,
uint8_t *ou, uint8_t *ov, int ouv_stride,
int oheight, int owidth) {
vp9_resize_plane(y, height, width, y_stride,
oy, oheight, owidth, oy_stride);
vp9_resize_plane(u, height / 2, width / 2, uv_stride,
ou, oheight / 2, owidth / 2, ouv_stride);
vp9_resize_plane(v, height / 2, width / 2, uv_stride,
ov, oheight / 2, owidth / 2, ouv_stride);
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth) {
vp9_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
vp9_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
owidth / 2, ouv_stride);
vp9_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
owidth / 2, ouv_stride);
}
void vp9_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride,
int height, int width,
uint8_t *oy, int oy_stride,
uint8_t *ou, uint8_t *ov, int ouv_stride,
int oheight, int owidth) {
vp9_resize_plane(y, height, width, y_stride,
oy, oheight, owidth, oy_stride);
vp9_resize_plane(u, height, width / 2, uv_stride,
ou, oheight, owidth / 2, ouv_stride);
vp9_resize_plane(v, height, width / 2, uv_stride,
ov, oheight, owidth / 2, ouv_stride);
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth) {
vp9_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
vp9_resize_plane(u, height, width / 2, uv_stride, ou, oheight, owidth / 2,
ouv_stride);
vp9_resize_plane(v, height, width / 2, uv_stride, ov, oheight, owidth / 2,
ouv_stride);
}
void vp9_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride,
int height, int width,
uint8_t *oy, int oy_stride,
uint8_t *ou, uint8_t *ov, int ouv_stride,
int oheight, int owidth) {
vp9_resize_plane(y, height, width, y_stride,
oy, oheight, owidth, oy_stride);
vp9_resize_plane(u, height, width, uv_stride,
ou, oheight, owidth, ouv_stride);
vp9_resize_plane(v, height, width, uv_stride,
ov, oheight, owidth, ouv_stride);
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth) {
vp9_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
vp9_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
ouv_stride);
vp9_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
ouv_stride);
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_resize_frame420(const uint8_t *const y,
int y_stride,
void vp9_highbd_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride,
int height, int width,
uint8_t *oy, int oy_stride,
uint8_t *ou, uint8_t *ov, int ouv_stride,
int oheight, int owidth, int bd) {
vp9_highbd_resize_plane(y, height, width, y_stride,
oy, oheight, owidth, oy_stride, bd);
vp9_highbd_resize_plane(u, height / 2, width / 2, uv_stride,
ou, oheight / 2, owidth / 2, ouv_stride, bd);
vp9_highbd_resize_plane(v, height / 2, width / 2, uv_stride,
ov, oheight / 2, owidth / 2, ouv_stride, bd);
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd) {
vp9_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
oy_stride, bd);
vp9_highbd_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
owidth / 2, ouv_stride, bd);
vp9_highbd_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
owidth / 2, ouv_stride, bd);
}
void vp9_highbd_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride,
int height, int width,
uint8_t *oy, int oy_stride,
uint8_t *ou, uint8_t *ov, int ouv_stride,
int oheight, int owidth, int bd) {
vp9_highbd_resize_plane(y, height, width, y_stride,
oy, oheight, owidth, oy_stride, bd);
vp9_highbd_resize_plane(u, height, width / 2, uv_stride,
ou, oheight, owidth / 2, ouv_stride, bd);
vp9_highbd_resize_plane(v, height, width / 2, uv_stride,
ov, oheight, owidth / 2, ouv_stride, bd);
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd) {
vp9_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
oy_stride, bd);
vp9_highbd_resize_plane(u, height, width / 2, uv_stride, ou, oheight,
owidth / 2, ouv_stride, bd);
vp9_highbd_resize_plane(v, height, width / 2, uv_stride, ov, oheight,
owidth / 2, ouv_stride, bd);
}
void vp9_highbd_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride,
int height, int width,
uint8_t *oy, int oy_stride,
uint8_t *ou, uint8_t *ov, int ouv_stride,
int oheight, int owidth, int bd) {
vp9_highbd_resize_plane(y, height, width, y_stride,
oy, oheight, owidth, oy_stride, bd);
vp9_highbd_resize_plane(u, height, width, uv_stride,
ou, oheight, owidth, ouv_stride, bd);
vp9_highbd_resize_plane(v, height, width, uv_stride,
ov, oheight, owidth, ouv_stride, bd);
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd) {
vp9_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
oy_stride, bd);
vp9_highbd_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
ouv_stride, bd);
vp9_highbd_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
ouv_stride, bd);
}
#endif // CONFIG_VP9_HIGHBITDEPTH

145
vp9/encoder/vp9_resize.h
View File

@@ -18,116 +18,51 @@
extern "C" {
#endif
void vp9_resize_plane(const uint8_t *const input,
int height,
int width,
int in_stride,
uint8_t *output,
int height2,
int width2,
void vp9_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2, int width2,
int out_stride);
void vp9_resize_frame420(const uint8_t *const y,
int y_stride,
const uint8_t *const u,
const uint8_t *const v,
int uv_stride,
int height,
int width,
uint8_t *oy,
int oy_stride,
uint8_t *ou,
uint8_t *ov,
int ouv_stride,
int oheight,
int owidth);
void vp9_resize_frame422(const uint8_t *const y,
int y_stride,
const uint8_t *const u,
const uint8_t *const v,
int uv_stride,
int height,
int width,
uint8_t *oy,
int oy_stride,
uint8_t *ou,
uint8_t *ov,
int ouv_stride,
int oheight,
int owidth);
void vp9_resize_frame444(const uint8_t *const y,
int y_stride,
const uint8_t *const u,
const uint8_t *const v,
int uv_stride,
int height,
int width,
uint8_t *oy,
int oy_stride,
uint8_t *ou,
uint8_t *ov,
int ouv_stride,
int oheight,
int owidth);
void vp9_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth);
void vp9_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth);
void vp9_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width, uint8_t *oy,
int oy_stride, uint8_t *ou, uint8_t *ov,
int ouv_stride, int oheight, int owidth);
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_resize_plane(const uint8_t *const input,
int height,
int width,
int in_stride,
uint8_t *output,
int height2,
int width2,
int out_stride,
int bd);
void vp9_highbd_resize_frame420(const uint8_t *const y,
int y_stride,
const uint8_t *const u,
const uint8_t *const v,
int uv_stride,
int height,
int width,
uint8_t *oy,
int oy_stride,
uint8_t *ou,
uint8_t *ov,
int ouv_stride,
int oheight,
int owidth,
int bd);
void vp9_highbd_resize_frame422(const uint8_t *const y,
int y_stride,
const uint8_t *const u,
const uint8_t *const v,
int uv_stride,
int height,
int width,
uint8_t *oy,
int oy_stride,
uint8_t *ou,
uint8_t *ov,
int ouv_stride,
int oheight,
int owidth,
int bd);
void vp9_highbd_resize_frame444(const uint8_t *const y,
int y_stride,
const uint8_t *const u,
const uint8_t *const v,
int uv_stride,
int height,
int width,
uint8_t *oy,
int oy_stride,
uint8_t *ou,
uint8_t *ov,
int ouv_stride,
int oheight,
int owidth,
int bd);
#endif // CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_resize_plane(const uint8_t *const input, int height, int width,
int in_stride, uint8_t *output, int height2,
int width2, int out_stride, int bd);
void vp9_highbd_resize_frame420(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd);
void vp9_highbd_resize_frame422(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd);
void vp9_highbd_resize_frame444(const uint8_t *const y, int y_stride,
const uint8_t *const u, const uint8_t *const v,
int uv_stride, int height, int width,
uint8_t *oy, int oy_stride, uint8_t *ou,
uint8_t *ov, int ouv_stride, int oheight,
int owidth, int bd);
#endif // CONFIG_VP9_HIGHBITDEPTH
#ifdef __cplusplus
} // extern "C"
#endif
#endif // VP9_ENCODER_VP9_RESIZE_H_
#endif // VP9_ENCODER_VP9_RESIZE_H_

48
vp9/encoder/vp9_segmentation.c
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include <limits.h>
#include "vpx_mem/vpx_mem.h"
@@ -31,8 +30,7 @@ void vp9_disable_segmentation(struct segmentation *seg) {
seg->update_data = 0;
}
void vp9_set_segment_data(struct segmentation *seg,
signed char *feature_data,
void vp9_set_segment_data(struct segmentation *seg, signed char *feature_data,
unsigned char abs_delta) {
seg->abs_delta = abs_delta;
@@ -75,13 +73,11 @@ static int cost_segmap(int *segcounts, vpx_prob *probs) {
const int c4567 = c45 + c67;
// Cost the top node of the tree
int cost = c0123 * vp9_cost_zero(probs[0]) +
c4567 * vp9_cost_one(probs[0]);
int cost = c0123 * vp9_cost_zero(probs[0]) + c4567 * vp9_cost_one(probs[0]);
// Cost subsequent levels
if (c0123 > 0) {
cost += c01 * vp9_cost_zero(probs[1]) +
c23 * vp9_cost_one(probs[1]);
cost += c01 * vp9_cost_zero(probs[1]) + c23 * vp9_cost_one(probs[1]);
if (c01 > 0)
cost += segcounts[0] * vp9_cost_zero(probs[3]) +
@@ -92,8 +88,7 @@ static int cost_segmap(int *segcounts, vpx_prob *probs) {
}
if (c4567 > 0) {
cost += c45 * vp9_cost_zero(probs[2]) +
c67 * vp9_cost_one(probs[2]);
cost += c45 * vp9_cost_zero(probs[2]) + c67 * vp9_cost_one(probs[2]);
if (c45 > 0)
cost += segcounts[4] * vp9_cost_zero(probs[5]) +
@@ -110,12 +105,11 @@ static void count_segs(const VP9_COMMON *cm, MACROBLOCKD *xd,
const TileInfo *tile, MODE_INFO **mi,
int *no_pred_segcounts,
int (*temporal_predictor_count)[2],
int *t_unpred_seg_counts,
int bw, int bh, int mi_row, int mi_col) {
int *t_unpred_seg_counts, int bw, int bh, int mi_row,
int mi_col) {
int segment_id;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
return;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
xd->mi = mi;
segment_id = xd->mi[0]->segment_id;
@@ -129,8 +123,8 @@ static void count_segs(const VP9_COMMON *cm, MACROBLOCKD *xd,
if (cm->frame_type != KEY_FRAME) {
const BLOCK_SIZE bsize = xd->mi[0]->sb_type;
// Test to see if the segment id matches the predicted value.
const int pred_segment_id = get_segment_id(cm, cm->last_frame_seg_map,
bsize, mi_row, mi_col);
const int pred_segment_id =
get_segment_id(cm, cm->last_frame_seg_map, bsize, mi_row, mi_col);
const int pred_flag = pred_segment_id == segment_id;
const int pred_context = vp9_get_pred_context_seg_id(xd);
@@ -140,8 +134,7 @@ static void count_segs(const VP9_COMMON *cm, MACROBLOCKD *xd,
temporal_predictor_count[pred_context][pred_flag]++;
// Update the "unpredicted" segment count
if (!pred_flag)
t_unpred_seg_counts[segment_id]++;
if (!pred_flag) t_unpred_seg_counts[segment_id]++;
}
}
@@ -149,15 +142,13 @@ static void count_segs_sb(const VP9_COMMON *cm, MACROBLOCKD *xd,
const TileInfo *tile, MODE_INFO **mi,
int *no_pred_segcounts,
int (*temporal_predictor_count)[2],
int *t_unpred_seg_counts,
int mi_row, int mi_col,
int *t_unpred_seg_counts, int mi_row, int mi_col,
BLOCK_SIZE bsize) {
const int mis = cm->mi_stride;
int bw, bh;
const int bs = num_8x8_blocks_wide_lookup[bsize], hbs = bs / 2;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
return;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols) return;
bw = num_8x8_blocks_wide_lookup[mi[0]->sb_type];
bh = num_8x8_blocks_high_lookup[mi[0]->sb_type];
@@ -174,9 +165,9 @@ static void count_segs_sb(const VP9_COMMON *cm, MACROBLOCKD *xd,
} else if (bw < bs && bh == bs) {
count_segs(cm, xd, tile, mi, no_pred_segcounts, temporal_predictor_count,
t_unpred_seg_counts, hbs, bs, mi_row, mi_col);
count_segs(cm, xd, tile, mi + hbs,
no_pred_segcounts, temporal_predictor_count, t_unpred_seg_counts,
hbs, bs, mi_row, mi_col + hbs);
count_segs(cm, xd, tile, mi + hbs, no_pred_segcounts,
temporal_predictor_count, t_unpred_seg_counts, hbs, bs, mi_row,
mi_col + hbs);
} else {
const BLOCK_SIZE subsize = subsize_lookup[PARTITION_SPLIT][bsize];
int n;
@@ -187,9 +178,8 @@ static void count_segs_sb(const VP9_COMMON *cm, MACROBLOCKD *xd,
const int mi_dc = hbs * (n & 1);
const int mi_dr = hbs * (n >> 1);
count_segs_sb(cm, xd, tile, &mi[mi_dr * mis + mi_dc],
no_pred_segcounts, temporal_predictor_count,
t_unpred_seg_counts,
count_segs_sb(cm, xd, tile, &mi[mi_dr * mis + mi_dc], no_pred_segcounts,
temporal_predictor_count, t_unpred_seg_counts,
mi_row + mi_dr, mi_col + mi_dc, subsize);
}
}
@@ -230,8 +220,8 @@ void vp9_choose_segmap_coding_method(VP9_COMMON *cm, MACROBLOCKD *xd) {
for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end;
mi_col += 8, mi += 8)
count_segs_sb(cm, xd, &tile, mi, no_pred_segcounts,
temporal_predictor_count, t_unpred_seg_counts,
mi_row, mi_col, BLOCK_64X64);
temporal_predictor_count, t_unpred_seg_counts, mi_row,
mi_col, BLOCK_64X64);
}
}

7
vp9/encoder/vp9_segmentation.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_SEGMENTATION_H_
#define VP9_ENCODER_VP9_SEGMENTATION_H_
@@ -22,11 +21,9 @@ extern "C" {
void vp9_enable_segmentation(struct segmentation *seg);
void vp9_disable_segmentation(struct segmentation *seg);
void vp9_disable_segfeature(struct segmentation *seg,
int segment_id,
void vp9_disable_segfeature(struct segmentation *seg, int segment_id,
SEG_LVL_FEATURES feature_id);
void vp9_clear_segdata(struct segmentation *seg,
int segment_id,
void vp9_clear_segdata(struct segmentation *seg, int segment_id,
SEG_LVL_FEATURES feature_id);
// The values given for each segment can be either deltas (from the default

79
vp9/encoder/vp9_skin_detection.c
View File

@@ -18,11 +18,14 @@
#define MODEL_MODE 1
// Fixed-point skin color model parameters.
static const int skin_mean[5][2] = {
{7463, 9614}, {6400, 10240}, {7040, 10240}, {8320, 9280}, {6800, 9614}};
static const int skin_inv_cov[4] = {4107, 1663, 1663, 2157}; // q16
static const int skin_threshold[6] = {1570636, 1400000, 800000, 800000, 800000,
800000}; // q18
static const int skin_mean[5][2] = { { 7463, 9614 },
{ 6400, 10240 },
{ 7040, 10240 },
{ 8320, 9280 },
{ 6800, 9614 } };
static const int skin_inv_cov[4] = { 4107, 1663, 1663, 2157 }; // q16
static const int skin_threshold[6] = { 1570636, 1400000, 800000,
800000, 800000, 800000 }; // q18
// Thresholds on luminance.
static const int y_low = 40;
@@ -41,10 +44,9 @@ static int evaluate_skin_color_difference(int cb, int cr, int idx) {
const int cb_diff_q2 = (cb_diff_q12 + (1 << 9)) >> 10;
const int cbcr_diff_q2 = (cbcr_diff_q12 + (1 << 9)) >> 10;
const int cr_diff_q2 = (cr_diff_q12 + (1 << 9)) >> 10;
const int skin_diff = skin_inv_cov[0] * cb_diff_q2 +
skin_inv_cov[1] * cbcr_diff_q2 +
skin_inv_cov[2] * cbcr_diff_q2 +
skin_inv_cov[3] * cr_diff_q2;
const int skin_diff =
skin_inv_cov[0] * cb_diff_q2 + skin_inv_cov[1] * cbcr_diff_q2 +
skin_inv_cov[2] * cbcr_diff_q2 + skin_inv_cov[3] * cr_diff_q2;
return skin_diff;
}
@@ -58,20 +60,18 @@ int vp9_skin_pixel(const uint8_t y, const uint8_t cb, const uint8_t cr,
} else {
int i = 0;
// Exit on grey.
if (cb == 128 && cr == 128)
return 0;
if (cb == 128 && cr == 128) return 0;
// Exit on very strong cb.
if (cb > 150 && cr < 110)
return 0;
if (cb > 150 && cr < 110) return 0;
for (; i < 5; i++) {
int skin_color_diff = evaluate_skin_color_difference(cb, cr, i);
if (skin_color_diff < skin_threshold[i + 1]) {
if (y < 60 && skin_color_diff > 3 * (skin_threshold[i + 1] >> 2))
return 0;
else if (motion == 0 &&
skin_color_diff > (skin_threshold[i + 1] >> 1))
return 0;
else
if (y < 60 && skin_color_diff > 3 * (skin_threshold[i + 1] >> 2))
return 0;
else if (motion == 0 &&
skin_color_diff > (skin_threshold[i + 1] >> 1))
return 0;
else
return 1;
}
// Exit if difference is much large than the threshold.
@@ -100,13 +100,11 @@ int vp9_compute_skin_block(const uint8_t *y, const uint8_t *u, const uint8_t *v,
const uint8_t ysource = y[y_height_shift * stride + y_width_shift];
const uint8_t usource = u[uv_height_shift * strideuv + uv_width_shift];
const uint8_t vsource = v[uv_height_shift * strideuv + uv_width_shift];
if (consec_zeromv > 25 && curr_motion_magn == 0)
motion = 0;
if (consec_zeromv > 25 && curr_motion_magn == 0) motion = 0;
return vp9_skin_pixel(ysource, usource, vsource, motion);
}
}
#ifdef OUTPUT_YUV_SKINMAP
// For viewing skin map on input source.
void vp9_compute_skin_map(VP9_COMP *const cpi, FILE *yuv_skinmap_file) {
@@ -129,11 +127,11 @@ void vp9_compute_skin_map(VP9_COMP *const cpi, FILE *yuv_skinmap_file) {
int mode_filter = 0;
YV12_BUFFER_CONFIG skinmap;
memset(&skinmap, 0, sizeof(YV12_BUFFER_CONFIG));
if (vpx_alloc_frame_buffer(&skinmap, cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
VP9_ENC_BORDER_IN_PIXELS, cm->byte_alignment)) {
vpx_free_frame_buffer(&skinmap);
return;
if (vpx_alloc_frame_buffer(&skinmap, cm->width, cm->height, cm->subsampling_x,
cm->subsampling_y, VP9_ENC_BORDER_IN_PIXELS,
cm->byte_alignment)) {
vpx_free_frame_buffer(&skinmap);
return;
}
memset(skinmap.buffer_alloc, 128, skinmap.frame_size);
y = skinmap.y_buffer;
@@ -153,11 +151,11 @@ void vp9_compute_skin_map(VP9_COMP *const cpi, FILE *yuv_skinmap_file) {
uint8_t usource2 = src_u[(uvpos + 1) * src_uvstride + uvpos];
uint8_t vsource2 = src_v[(uvpos + 1) * src_uvstride + uvpos];
uint8_t ysource3 = src_y[ypos * src_ystride + (ypos + 1)];
uint8_t usource3 = src_u[uvpos * src_uvstride + (uvpos + 1)];
uint8_t vsource3 = src_v[uvpos * src_uvstride + (uvpos + 1)];
uint8_t usource3 = src_u[uvpos * src_uvstride + (uvpos + 1)];
uint8_t vsource3 = src_v[uvpos * src_uvstride + (uvpos + 1)];
uint8_t ysource4 = src_y[(ypos + 1) * src_ystride + (ypos + 1)];
uint8_t usource4 = src_u[(uvpos + 1) * src_uvstride + (uvpos + 1)];
uint8_t vsource4 = src_v[(uvpos + 1) * src_uvstride + (uvpos + 1)];
uint8_t usource4 = src_u[(uvpos + 1) * src_uvstride + (uvpos + 1)];
uint8_t vsource4 = src_v[(uvpos + 1) * src_uvstride + (uvpos + 1)];
ysource = (ysource + ysource2 + ysource3 + ysource4) >> 2;
usource = (usource + usource2 + usource3 + usource4) >> 2;
vsource = (vsource + vsource2 + vsource3 + vsource4) >> 2;
@@ -172,16 +170,15 @@ void vp9_compute_skin_map(VP9_COMP *const cpi, FILE *yuv_skinmap_file) {
if (y_bsize == 8)
consec_zeromv = cpi->consec_zero_mv[bl_index];
else
consec_zeromv = VPXMIN(cpi->consec_zero_mv[bl_index],
VPXMIN(cpi->consec_zero_mv[bl_index1],
VPXMIN(cpi->consec_zero_mv[bl_index2],
cpi->consec_zero_mv[bl_index3])));
if (y_bsize == 16)
block_size = BLOCK_16X16;
is_skin = vp9_compute_skin_block(src_y, src_u, src_v, src_ystride,
src_uvstride, block_size,
consec_zeromv,
0);
consec_zeromv =
VPXMIN(cpi->consec_zero_mv[bl_index],
VPXMIN(cpi->consec_zero_mv[bl_index1],
VPXMIN(cpi->consec_zero_mv[bl_index2],
cpi->consec_zero_mv[bl_index3])));
if (y_bsize == 16) block_size = BLOCK_16X16;
is_skin =
vp9_compute_skin_block(src_y, src_u, src_v, src_ystride,
src_uvstride, block_size, consec_zeromv, 0);
}
for (i = 0; i < y_bsize; i++) {
for (j = 0; j < y_bsize; j++) {

117
vp9/encoder/vp9_speed_features.c
View File

@@ -16,21 +16,23 @@
#include "vpx_dsp/vpx_dsp_common.h"
// Mesh search patters for various speed settings
static MESH_PATTERN best_quality_mesh_pattern[MAX_MESH_STEP] =
{{64, 4}, {28, 2}, {15, 1}, {7, 1}};
static MESH_PATTERN best_quality_mesh_pattern[MAX_MESH_STEP] = {
{ 64, 4 }, { 28, 2 }, { 15, 1 }, { 7, 1 }
};
#define MAX_MESH_SPEED 5 // Max speed setting for mesh motion method
static MESH_PATTERN good_quality_mesh_patterns[MAX_MESH_SPEED + 1]
[MAX_MESH_STEP] =
{{{64, 8}, {28, 4}, {15, 1}, {7, 1}},
{{64, 8}, {28, 4}, {15, 1}, {7, 1}},
{{64, 8}, {14, 2}, {7, 1}, {7, 1}},
{{64, 16}, {24, 8}, {12, 4}, {7, 1}},
{{64, 16}, {24, 8}, {12, 4}, {7, 1}},
{{64, 16}, {24, 8}, {12, 4}, {7, 1}},
static MESH_PATTERN
good_quality_mesh_patterns[MAX_MESH_SPEED + 1][MAX_MESH_STEP] = {
{ { 64, 8 }, { 28, 4 }, { 15, 1 }, { 7, 1 } },
{ { 64, 8 }, { 28, 4 }, { 15, 1 }, { 7, 1 } },
{ { 64, 8 }, { 14, 2 }, { 7, 1 }, { 7, 1 } },
{ { 64, 16 }, { 24, 8 }, { 12, 4 }, { 7, 1 } },
{ { 64, 16 }, { 24, 8 }, { 12, 4 }, { 7, 1 } },
{ { 64, 16 }, { 24, 8 }, { 12, 4 }, { 7, 1 } },
};
static unsigned char good_quality_max_mesh_pct[MAX_MESH_SPEED + 1] =
{50, 25, 15, 5, 1, 1};
static unsigned char good_quality_max_mesh_pct[MAX_MESH_SPEED + 1] = {
50, 25, 15, 5, 1, 1
};
// Intra only frames, golden frames (except alt ref overlays) and
// alt ref frames tend to be coded at a higher than ambient quality
@@ -67,8 +69,8 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
if (speed >= 1) {
if (VPXMIN(cm->width, cm->height) >= 720) {
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
sf->disable_split_mask =
cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
sf->partition_search_breakout_dist_thr = (1 << 23);
} else {
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
@@ -78,8 +80,8 @@ static void set_good_speed_feature_framesize_dependent(VP9_COMP *cpi,
if (speed >= 2) {
if (VPXMIN(cm->width, cm->height) >= 720) {
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
sf->disable_split_mask =
cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
sf->adaptive_pred_interp_filter = 0;
sf->partition_search_breakout_dist_thr = (1 << 24);
sf->partition_search_breakout_rate_thr = 120;
@@ -147,7 +149,7 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
}
sf->use_square_only_threshold = BLOCK_4X4;
sf->less_rectangular_check = 1;
sf->less_rectangular_check = 1;
sf->use_rd_breakout = 1;
sf->adaptive_motion_search = 1;
@@ -167,17 +169,17 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
}
if (speed >= 2) {
sf->tx_size_search_method = frame_is_boosted(cpi) ? USE_FULL_RD
: USE_LARGESTALL;
sf->tx_size_search_method =
frame_is_boosted(cpi) ? USE_FULL_RD : USE_LARGESTALL;
// Reference masking is not supported in dynamic scaling mode.
sf->reference_masking = cpi->oxcf.resize_mode != RESIZE_DYNAMIC ? 1 : 0;
sf->mode_search_skip_flags = (cm->frame_type == KEY_FRAME) ? 0 :
FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->mode_search_skip_flags =
(cm->frame_type == KEY_FRAME) ? 0 : FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->disable_filter_search_var_thresh = 100;
sf->comp_inter_joint_search_thresh = BLOCK_SIZES;
sf->auto_min_max_partition_size = RELAXED_NEIGHBORING_MIN_MAX;
@@ -186,8 +188,8 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
if (speed >= 3) {
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->tx_size_search_method = frame_is_intra_only(cm) ? USE_FULL_RD
: USE_LARGESTALL;
sf->tx_size_search_method =
frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL;
sf->mv.subpel_search_method = SUBPEL_TREE_PRUNED;
sf->adaptive_pred_interp_filter = 0;
sf->adaptive_mode_search = 1;
@@ -234,13 +236,14 @@ static void set_good_speed_feature(VP9_COMP *cpi, VP9_COMMON *cm,
}
static void set_rt_speed_feature_framesize_dependent(VP9_COMP *cpi,
SPEED_FEATURES *sf, int speed) {
SPEED_FEATURES *sf,
int speed) {
VP9_COMMON *const cm = &cpi->common;
if (speed >= 1) {
if (VPXMIN(cm->width, cm->height) >= 720) {
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
sf->disable_split_mask =
cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
} else {
sf->disable_split_mask = DISABLE_COMPOUND_SPLIT;
}
@@ -248,8 +251,8 @@ static void set_rt_speed_feature_framesize_dependent(VP9_COMP *cpi,
if (speed >= 2) {
if (VPXMIN(cm->width, cm->height) >= 720) {
sf->disable_split_mask = cm->show_frame ? DISABLE_ALL_SPLIT
: DISABLE_ALL_INTER_SPLIT;
sf->disable_split_mask =
cm->show_frame ? DISABLE_ALL_SPLIT : DISABLE_ALL_INTER_SPLIT;
} else {
sf->disable_split_mask = LAST_AND_INTRA_SPLIT_ONLY;
}
@@ -264,13 +267,13 @@ static void set_rt_speed_feature_framesize_dependent(VP9_COMP *cpi,
}
if (speed >= 7) {
sf->encode_breakout_thresh = (VPXMIN(cm->width, cm->height) >= 720) ?
800 : 300;
sf->encode_breakout_thresh =
(VPXMIN(cm->width, cm->height) >= 720) ? 800 : 300;
}
}
static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
int speed, vp9e_tune_content content) {
static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf, int speed,
vp9e_tune_content content) {
VP9_COMMON *const cm = &cpi->common;
const int is_keyframe = cm->frame_type == KEY_FRAME;
const int frames_since_key = is_keyframe ? 0 : cpi->rc.frames_since_key;
@@ -285,8 +288,8 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
sf->quant_coeff_opt = 0;
sf->use_square_partition_only = !frame_is_intra_only(cm);
sf->less_rectangular_check = 1;
sf->tx_size_search_method = frame_is_intra_only(cm) ? USE_FULL_RD
: USE_LARGESTALL;
sf->tx_size_search_method =
frame_is_intra_only(cm) ? USE_FULL_RD : USE_LARGESTALL;
sf->use_rd_breakout = 1;
@@ -300,11 +303,11 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
}
if (speed >= 2) {
sf->mode_search_skip_flags = (cm->frame_type == KEY_FRAME) ? 0 :
FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->mode_search_skip_flags =
(cm->frame_type == KEY_FRAME) ? 0 : FLAG_SKIP_INTRA_DIRMISMATCH |
FLAG_SKIP_INTRA_BESTINTER |
FLAG_SKIP_COMP_BESTINTRA |
FLAG_SKIP_INTRA_LOWVAR;
sf->adaptive_pred_interp_filter = 2;
// Reference masking only enabled for 1 spatial layer, and if none of the
@@ -315,15 +318,14 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
(cpi->external_resize == 1 ||
cpi->oxcf.resize_mode == RESIZE_DYNAMIC)) {
MV_REFERENCE_FRAME ref_frame;
static const int flag_list[4] =
{0, VP9_LAST_FLAG, VP9_GOLD_FLAG, VP9_ALT_FLAG};
static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG,
VP9_ALT_FLAG };
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame);
if (yv12 != NULL && (cpi->ref_frame_flags & flag_list[ref_frame])) {
const struct scale_factors *const scale_fac =
&cm->frame_refs[ref_frame - 1].sf;
if (vp9_is_scaled(scale_fac))
sf->reference_masking = 0;
if (vp9_is_scaled(scale_fac)) sf->reference_masking = 0;
}
}
}
@@ -360,8 +362,8 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
sf->use_fast_coef_costing = 0;
sf->auto_min_max_partition_size = STRICT_NEIGHBORING_MIN_MAX;
sf->adjust_partitioning_from_last_frame =
cm->last_frame_type != cm->frame_type || (0 ==
(frames_since_key + 1) % sf->last_partitioning_redo_frequency);
cm->last_frame_type != cm->frame_type ||
(0 == (frames_since_key + 1) % sf->last_partitioning_redo_frequency);
sf->mv.subpel_force_stop = 1;
for (i = 0; i < TX_SIZES; i++) {
sf->intra_y_mode_mask[i] = INTRA_DC_H_V;
@@ -381,11 +383,12 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
if (speed >= 5) {
sf->use_quant_fp = !is_keyframe;
sf->auto_min_max_partition_size = is_keyframe ? RELAXED_NEIGHBORING_MIN_MAX
: STRICT_NEIGHBORING_MIN_MAX;
sf->auto_min_max_partition_size =
is_keyframe ? RELAXED_NEIGHBORING_MIN_MAX : STRICT_NEIGHBORING_MIN_MAX;
sf->default_max_partition_size = BLOCK_32X32;
sf->default_min_partition_size = BLOCK_8X8;
sf->force_frame_boost = is_keyframe ||
sf->force_frame_boost =
is_keyframe ||
(frames_since_key % (sf->last_partitioning_redo_frequency << 1) == 1);
sf->max_delta_qindex = is_keyframe ? 20 : 15;
sf->partition_search_type = REFERENCE_PARTITION;
@@ -404,8 +407,7 @@ static void set_rt_speed_feature(VP9_COMP *cpi, SPEED_FEATURES *sf,
sf->mode_search_skip_flags = FLAG_SKIP_INTRA_DIRMISMATCH;
sf->tx_size_search_method = is_keyframe ? USE_LARGESTALL : USE_TX_8X8;
sf->simple_model_rd_from_var = 1;
if (cpi->oxcf.rc_mode == VPX_VBR)
sf->mv.search_method = NSTEP;
if (cpi->oxcf.rc_mode == VPX_VBR) sf->mv.search_method = NSTEP;
if (!is_keyframe) {
int i;
@@ -562,8 +564,7 @@ void vp9_set_speed_features_framesize_independent(VP9_COMP *cpi) {
sf->mode_skip_start = MAX_MODES; // Mode index at which mode skip mask set
sf->schedule_mode_search = 0;
sf->use_nonrd_pick_mode = 0;
for (i = 0; i < BLOCK_SIZES; ++i)
sf->inter_mode_mask[i] = INTER_ALL;
for (i = 0; i < BLOCK_SIZES; ++i) sf->inter_mode_mask[i] = INTER_ALL;
sf->max_intra_bsize = BLOCK_64X64;
sf->reuse_inter_pred_sby = 0;
// This setting only takes effect when partition_search_type is set
@@ -614,8 +615,7 @@ void vp9_set_speed_features_framesize_independent(VP9_COMP *cpi) {
sf->exhaustive_searches_thresh = sf->exhaustive_searches_thresh << 1;
for (i = 0; i < MAX_MESH_STEP; ++i) {
sf->mesh_patterns[i].range =
good_quality_mesh_patterns[speed][i].range;
sf->mesh_patterns[i].range = good_quality_mesh_patterns[speed][i].range;
sf->mesh_patterns[i].interval =
good_quality_mesh_patterns[speed][i].interval;
}
@@ -623,8 +623,7 @@ void vp9_set_speed_features_framesize_independent(VP9_COMP *cpi) {
// Slow quant, dct and trellis not worthwhile for first pass
// so make sure they are always turned off.
if (oxcf->pass == 1)
sf->optimize_coefficients = 0;
if (oxcf->pass == 1) sf->optimize_coefficients = 0;
// No recode for 1 pass.
if (oxcf->pass == 0) {

36
vp9/encoder/vp9_speed_features.h
View File

@@ -18,17 +18,14 @@ extern "C" {
#endif
enum {
INTRA_ALL = (1 << DC_PRED) |
(1 << V_PRED) | (1 << H_PRED) |
(1 << D45_PRED) | (1 << D135_PRED) |
(1 << D117_PRED) | (1 << D153_PRED) |
(1 << D207_PRED) | (1 << D63_PRED) |
(1 << TM_PRED),
INTRA_DC = (1 << DC_PRED),
INTRA_DC_TM = (1 << DC_PRED) | (1 << TM_PRED),
INTRA_DC_H_V = (1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED),
INTRA_DC_TM_H_V = (1 << DC_PRED) | (1 << TM_PRED) | (1 << V_PRED) |
(1 << H_PRED)
INTRA_ALL = (1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED) | (1 << D45_PRED) |
(1 << D135_PRED) | (1 << D117_PRED) | (1 << D153_PRED) |
(1 << D207_PRED) | (1 << D63_PRED) | (1 << TM_PRED),
INTRA_DC = (1 << DC_PRED),
INTRA_DC_TM = (1 << DC_PRED) | (1 << TM_PRED),
INTRA_DC_H_V = (1 << DC_PRED) | (1 << V_PRED) | (1 << H_PRED),
INTRA_DC_TM_H_V =
(1 << DC_PRED) | (1 << TM_PRED) | (1 << V_PRED) | (1 << H_PRED)
};
enum {
@@ -42,20 +39,15 @@ enum {
};
enum {
DISABLE_ALL_INTER_SPLIT = (1 << THR_COMP_GA) |
(1 << THR_COMP_LA) |
(1 << THR_ALTR) |
(1 << THR_GOLD) |
(1 << THR_LAST),
DISABLE_ALL_INTER_SPLIT = (1 << THR_COMP_GA) | (1 << THR_COMP_LA) |
(1 << THR_ALTR) | (1 << THR_GOLD) | (1 << THR_LAST),
DISABLE_ALL_SPLIT = (1 << THR_INTRA) | DISABLE_ALL_INTER_SPLIT,
DISABLE_ALL_SPLIT = (1 << THR_INTRA) | DISABLE_ALL_INTER_SPLIT,
DISABLE_COMPOUND_SPLIT = (1 << THR_COMP_GA) | (1 << THR_COMP_LA),
DISABLE_COMPOUND_SPLIT = (1 << THR_COMP_GA) | (1 << THR_COMP_LA),
LAST_AND_INTRA_SPLIT_ONLY = (1 << THR_COMP_GA) |
(1 << THR_COMP_LA) |
(1 << THR_ALTR) |
(1 << THR_GOLD)
LAST_AND_INTRA_SPLIT_ONLY = (1 << THR_COMP_GA) | (1 << THR_COMP_LA) |
(1 << THR_ALTR) | (1 << THR_GOLD)
};
typedef enum {

76
vp9/encoder/vp9_subexp.c
View File

@@ -15,22 +15,20 @@
#include "vp9/encoder/vp9_subexp.h"
static const uint8_t update_bits[255] = {
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 0,
5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 6, 6, 6,
6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
8, 8, 8, 8, 8, 8, 8, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
11, 11, 11, 11, 11, 11, 11, 0,
};
#define MIN_DELP_BITS 5
@@ -48,23 +46,23 @@ static int remap_prob(int v, int m) {
static const uint8_t map_table[MAX_PROB - 1] = {
// generated by:
// map_table[j] = split_index(j, MAX_PROB - 1, MODULUS_PARAM);
20, 21, 22, 23, 24, 25, 0, 26, 27, 28, 29, 30, 31, 32, 33,
34, 35, 36, 37, 1, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47,
48, 49, 2, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61,
3, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 4, 74,
75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 5, 86, 87, 88,
89, 90, 91, 92, 93, 94, 95, 96, 97, 6, 98, 99, 100, 101, 102,
103, 104, 105, 106, 107, 108, 109, 7, 110, 111, 112, 113, 114, 115, 116,
117, 118, 119, 120, 121, 8, 122, 123, 124, 125, 126, 127, 128, 129, 130,
131, 132, 133, 9, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144,
145, 10, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 11,
158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 12, 170, 171,
172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 13, 182, 183, 184, 185,
186, 187, 188, 189, 190, 191, 192, 193, 14, 194, 195, 196, 197, 198, 199,
200, 201, 202, 203, 204, 205, 15, 206, 207, 208, 209, 210, 211, 212, 213,
214, 215, 216, 217, 16, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227,
228, 229, 17, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241,
18, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 19,
20, 21, 22, 23, 24, 25, 0, 26, 27, 28, 29, 30, 31, 32, 33,
34, 35, 36, 37, 1, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47,
48, 49, 2, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61,
3, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 4, 74,
75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 5, 86, 87, 88,
89, 90, 91, 92, 93, 94, 95, 96, 97, 6, 98, 99, 100, 101, 102,
103, 104, 105, 106, 107, 108, 109, 7, 110, 111, 112, 113, 114, 115, 116,
117, 118, 119, 120, 121, 8, 122, 123, 124, 125, 126, 127, 128, 129, 130,
131, 132, 133, 9, 134, 135, 136, 137, 138, 139, 140, 141, 142, 143, 144,
145, 10, 146, 147, 148, 149, 150, 151, 152, 153, 154, 155, 156, 157, 11,
158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 12, 170, 171,
172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 13, 182, 183, 184, 185,
186, 187, 188, 189, 190, 191, 192, 193, 14, 194, 195, 196, 197, 198, 199,
200, 201, 202, 203, 204, 205, 15, 206, 207, 208, 209, 210, 211, 212, 213,
214, 215, 216, 217, 16, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227,
228, 229, 17, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241,
18, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 19,
};
v--;
m--;
@@ -115,9 +113,8 @@ void vp9_write_prob_diff_update(vpx_writer *w, vpx_prob newp, vpx_prob oldp) {
encode_term_subexp(w, delp);
}
int vp9_prob_diff_update_savings_search(const unsigned int *ct,
vpx_prob oldp, vpx_prob *bestp,
vpx_prob upd) {
int vp9_prob_diff_update_savings_search(const unsigned int *ct, vpx_prob oldp,
vpx_prob *bestp, vpx_prob upd) {
const int old_b = cost_branch256(ct, oldp);
int bestsavings = 0;
vpx_prob newp, bestnewp = oldp;
@@ -141,8 +138,7 @@ int vp9_prob_diff_update_savings_search(const unsigned int *ct,
int vp9_prob_diff_update_savings_search_model(const unsigned int *ct,
const vpx_prob oldp,
vpx_prob *bestp,
vpx_prob upd,
vpx_prob *bestp, vpx_prob upd,
int stepsize) {
int i, old_b, new_b, update_b, savings, bestsavings;
int newp;
@@ -185,8 +181,8 @@ void vp9_cond_prob_diff_update(vpx_writer *w, vpx_prob *oldp,
const unsigned int ct[2]) {
const vpx_prob upd = DIFF_UPDATE_PROB;
vpx_prob newp = get_binary_prob(ct[0], ct[1]);
const int savings = vp9_prob_diff_update_savings_search(ct, *oldp, &newp,
upd);
const int savings =
vp9_prob_diff_update_savings_search(ct, *oldp, &newp, upd);
assert(newp >= 1);
if (savings > 0) {
vpx_write(w, 1, upd);

14
vp9/encoder/vp9_subexp.h
View File

@@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VP9_ENCODER_VP9_SUBEXP_H_
#define VP9_ENCODER_VP9_SUBEXP_H_
@@ -20,21 +19,18 @@ extern "C" {
struct vpx_writer;
void vp9_write_prob_diff_update(struct vpx_writer *w,
vpx_prob newp, vpx_prob oldp);
void vp9_write_prob_diff_update(struct vpx_writer *w, vpx_prob newp,
vpx_prob oldp);
void vp9_cond_prob_diff_update(struct vpx_writer *w, vpx_prob *oldp,
const unsigned int ct[2]);
int vp9_prob_diff_update_savings_search(const unsigned int *ct,
vpx_prob oldp, vpx_prob *bestp,
vpx_prob upd);
int vp9_prob_diff_update_savings_search(const unsigned int *ct, vpx_prob oldp,
vpx_prob *bestp, vpx_prob upd);
int vp9_prob_diff_update_savings_search_model(const unsigned int *ct,
const vpx_prob oldp,
vpx_prob *bestp,
vpx_prob upd,
vpx_prob *bestp, vpx_prob upd,
int stepsize);
#ifdef __cplusplus

166
vp9/encoder/vp9_svc_layercontext.c
View File

@@ -16,7 +16,7 @@
#include "vp9/encoder/vp9_extend.h"
#include "vpx_dsp/vpx_dsp_common.h"
#define SMALL_FRAME_WIDTH 32
#define SMALL_FRAME_WIDTH 32
#define SMALL_FRAME_HEIGHT 16
void vp9_init_layer_context(VP9_COMP *const cpi) {
@@ -36,8 +36,7 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
svc->scaled_temp_is_alloc = 0;
svc->scaled_one_half = 0;
svc->current_superframe = 0;
for (i = 0; i < REF_FRAMES; ++i)
svc->ref_frame_index[i] = -1;
for (i = 0; i < REF_FRAMES; ++i) svc->ref_frame_index[i] = -1;
for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
cpi->svc.ext_frame_flags[sl] = 0;
cpi->svc.ext_lst_fb_idx[sl] = 0;
@@ -46,16 +45,14 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
}
if (cpi->oxcf.error_resilient_mode == 0 && cpi->oxcf.pass == 2) {
if (vpx_realloc_frame_buffer(&cpi->svc.empty_frame.img,
SMALL_FRAME_WIDTH, SMALL_FRAME_HEIGHT,
cpi->common.subsampling_x,
if (vpx_realloc_frame_buffer(&cpi->svc.empty_frame.img, SMALL_FRAME_WIDTH,
SMALL_FRAME_HEIGHT, cpi->common.subsampling_x,
cpi->common.subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cpi->common.use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS,
cpi->common.byte_alignment,
NULL, NULL, NULL))
cpi->common.byte_alignment, NULL, NULL, NULL))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate empty frame for multiple frame "
"contexts");
@@ -97,10 +94,10 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
lc->target_bandwidth = oxcf->layer_target_bitrate[layer];
lrc->last_q[KEY_FRAME] = oxcf->best_allowed_q;
lrc->last_q[INTER_FRAME] = oxcf->best_allowed_q;
lrc->avg_frame_qindex[KEY_FRAME] = (oxcf->worst_allowed_q +
oxcf->best_allowed_q) / 2;
lrc->avg_frame_qindex[INTER_FRAME] = (oxcf->worst_allowed_q +
oxcf->best_allowed_q) / 2;
lrc->avg_frame_qindex[KEY_FRAME] =
(oxcf->worst_allowed_q + oxcf->best_allowed_q) / 2;
lrc->avg_frame_qindex[INTER_FRAME] =
(oxcf->worst_allowed_q + oxcf->best_allowed_q) / 2;
if (oxcf->ss_enable_auto_arf[sl])
lc->alt_ref_idx = alt_ref_idx++;
else
@@ -108,16 +105,15 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
lc->gold_ref_idx = INVALID_IDX;
}
lrc->buffer_level = oxcf->starting_buffer_level_ms *
lc->target_bandwidth / 1000;
lrc->buffer_level =
oxcf->starting_buffer_level_ms * lc->target_bandwidth / 1000;
lrc->bits_off_target = lrc->buffer_level;
// Initialize the cyclic refresh parameters. If spatial layers are used
// (i.e., ss_number_layers > 1), these need to be updated per spatial
// layer.
// Cyclic refresh is only applied on base temporal layer.
if (oxcf->ss_number_layers > 1 &&
tl == 0) {
if (oxcf->ss_number_layers > 1 && tl == 0) {
size_t last_coded_q_map_size;
size_t consec_zero_mv_size;
VP9_COMMON *const cm = &cpi->common;
@@ -125,8 +121,8 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
CHECK_MEM_ERROR(cm, lc->map,
vpx_malloc(mi_rows * mi_cols * sizeof(*lc->map)));
memset(lc->map, 0, mi_rows * mi_cols);
last_coded_q_map_size = mi_rows * mi_cols *
sizeof(*lc->last_coded_q_map);
last_coded_q_map_size =
mi_rows * mi_cols * sizeof(*lc->last_coded_q_map);
CHECK_MEM_ERROR(cm, lc->last_coded_q_map,
vpx_malloc(last_coded_q_map_size));
assert(MAXQ <= 255);
@@ -140,8 +136,8 @@ void vp9_init_layer_context(VP9_COMP *const cpi) {
}
// Still have extra buffer for base layer golden frame
if (!(svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR)
&& alt_ref_idx < REF_FRAMES)
if (!(svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) &&
alt_ref_idx < REF_FRAMES)
svc->layer_context[0].gold_ref_idx = alt_ref_idx;
}
@@ -162,11 +158,12 @@ void vp9_update_layer_context_change_config(VP9_COMP *const cpi,
oxcf->layer_target_bitrate[layer];
}
layer = LAYER_IDS_TO_IDX(sl, ((oxcf->ts_number_layers - 1) < 0 ?
0 : (oxcf->ts_number_layers - 1)), oxcf->ts_number_layers);
spatial_layer_target =
svc->layer_context[layer].target_bandwidth =
oxcf->layer_target_bitrate[layer];
layer = LAYER_IDS_TO_IDX(
sl,
((oxcf->ts_number_layers - 1) < 0 ? 0 : (oxcf->ts_number_layers - 1)),
oxcf->ts_number_layers);
spatial_layer_target = svc->layer_context[layer].target_bandwidth =
oxcf->layer_target_bitrate[layer];
for (tl = 0; tl < oxcf->ts_number_layers; ++tl) {
LAYER_CONTEXT *const lc =
@@ -214,8 +211,8 @@ void vp9_update_layer_context_change_config(VP9_COMP *const cpi,
(int64_t)(rc->optimal_buffer_level * bitrate_alloc);
lrc->maximum_buffer_size =
(int64_t)(rc->maximum_buffer_size * bitrate_alloc);
lrc->bits_off_target = VPXMIN(lrc->bits_off_target,
lrc->maximum_buffer_size);
lrc->bits_off_target =
VPXMIN(lrc->bits_off_target, lrc->maximum_buffer_size);
lrc->buffer_level = VPXMIN(lrc->buffer_level, lrc->maximum_buffer_size);
// Update framerate-related quantities.
if (svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) {
@@ -235,12 +232,12 @@ void vp9_update_layer_context_change_config(VP9_COMP *const cpi,
static LAYER_CONTEXT *get_layer_context(VP9_COMP *const cpi) {
if (is_one_pass_cbr_svc(cpi))
return &cpi->svc.layer_context[cpi->svc.spatial_layer_id *
cpi->svc.number_temporal_layers + cpi->svc.temporal_layer_id];
cpi->svc.number_temporal_layers +
cpi->svc.temporal_layer_id];
else
return (cpi->svc.number_temporal_layers > 1 &&
cpi->oxcf.rc_mode == VPX_CBR) ?
&cpi->svc.layer_context[cpi->svc.temporal_layer_id] :
&cpi->svc.layer_context[cpi->svc.spatial_layer_id];
return (cpi->svc.number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR)
? &cpi->svc.layer_context[cpi->svc.temporal_layer_id]
: &cpi->svc.layer_context[cpi->svc.spatial_layer_id];
}
void vp9_update_temporal_layer_framerate(VP9_COMP *const cpi) {
@@ -250,7 +247,7 @@ void vp9_update_temporal_layer_framerate(VP9_COMP *const cpi) {
RATE_CONTROL *const lrc = &lc->rc;
// Index into spatial+temporal arrays.
const int st_idx = svc->spatial_layer_id * svc->number_temporal_layers +
svc->temporal_layer_id;
svc->temporal_layer_id;
const int tl = svc->temporal_layer_id;
lc->framerate = cpi->framerate / oxcf->ts_rate_decimator[tl];
@@ -277,10 +274,11 @@ void vp9_update_spatial_layer_framerate(VP9_COMP *const cpi, double framerate) {
lc->framerate = framerate;
lrc->avg_frame_bandwidth = (int)(lc->target_bandwidth / lc->framerate);
lrc->min_frame_bandwidth = (int)(lrc->avg_frame_bandwidth *
oxcf->two_pass_vbrmin_section / 100);
lrc->min_frame_bandwidth =
(int)(lrc->avg_frame_bandwidth * oxcf->two_pass_vbrmin_section / 100);
lrc->max_frame_bandwidth = (int)(((int64_t)lrc->avg_frame_bandwidth *
oxcf->two_pass_vbrmax_section) / 100);
oxcf->two_pass_vbrmax_section) /
100);
vp9_rc_set_gf_interval_range(cpi, lrc);
}
@@ -309,8 +307,7 @@ void vp9_restore_layer_context(VP9_COMP *const cpi) {
// For spatial-svc, allow cyclic-refresh to be applied on the spatial layers,
// for the base temporal layer.
if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
cpi->svc.number_spatial_layers > 1 &&
cpi->svc.temporal_layer_id == 0) {
cpi->svc.number_spatial_layers > 1 && cpi->svc.temporal_layer_id == 0) {
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
signed char *temp = cr->map;
uint8_t *temp2 = cr->last_coded_q_map;
@@ -337,8 +334,7 @@ void vp9_save_layer_context(VP9_COMP *const cpi) {
// For spatial-svc, allow cyclic-refresh to be applied on the spatial layers,
// for the base temporal layer.
if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ &&
cpi->svc.number_spatial_layers > 1 &&
cpi->svc.temporal_layer_id == 0) {
cpi->svc.number_spatial_layers > 1 && cpi->svc.temporal_layer_id == 0) {
CYCLIC_REFRESH *const cr = cpi->cyclic_refresh;
signed char *temp = lc->map;
uint8_t *temp2 = lc->last_coded_q_map;
@@ -380,20 +376,19 @@ void vp9_inc_frame_in_layer(VP9_COMP *const cpi) {
}
int vp9_is_upper_layer_key_frame(const VP9_COMP *const cpi) {
return is_two_pass_svc(cpi) &&
cpi->svc.spatial_layer_id > 0 &&
return is_two_pass_svc(cpi) && cpi->svc.spatial_layer_id > 0 &&
cpi->svc.layer_context[cpi->svc.spatial_layer_id *
cpi->svc.number_temporal_layers +
cpi->svc.temporal_layer_id].is_key_frame;
cpi->svc.number_temporal_layers +
cpi->svc.temporal_layer_id]
.is_key_frame;
}
static void get_layer_resolution(const int width_org, const int height_org,
const int num, const int den,
int *width_out, int *height_out) {
const int num, const int den, int *width_out,
int *height_out) {
int w, h;
if (width_out == NULL || height_out == NULL || den == 0)
return;
if (width_out == NULL || height_out == NULL || den == 0) return;
w = width_org * num / den;
h = height_org * num / den;
@@ -415,10 +410,13 @@ static void set_flags_and_fb_idx_for_temporal_mode3(VP9_COMP *const cpi) {
spatial_id = cpi->svc.spatial_layer_id = cpi->svc.spatial_layer_to_encode;
frame_num_within_temporal_struct =
cpi->svc.layer_context[cpi->svc.spatial_layer_id *
cpi->svc.number_temporal_layers].current_video_frame_in_layer % 4;
cpi->svc.number_temporal_layers]
.current_video_frame_in_layer %
4;
temporal_id = cpi->svc.temporal_layer_id =
(frame_num_within_temporal_struct & 1) ? 2 :
(frame_num_within_temporal_struct >> 1);
(frame_num_within_temporal_struct & 1)
? 2
: (frame_num_within_temporal_struct >> 1);
cpi->ext_refresh_last_frame = cpi->ext_refresh_golden_frame =
cpi->ext_refresh_alt_ref_frame = 0;
if (!temporal_id) {
@@ -465,7 +463,7 @@ static void set_flags_and_fb_idx_for_temporal_mode3(VP9_COMP *const cpi) {
if (spatial_id == cpi->svc.number_spatial_layers - 1) { // top layer
cpi->ext_refresh_frame_flags_pending = 1;
if (!spatial_id)
cpi->ref_frame_flags = VP9_LAST_FLAG;
cpi->ref_frame_flags = VP9_LAST_FLAG;
else
cpi->ref_frame_flags = VP9_LAST_FLAG | VP9_GOLD_FLAG;
} else if (!spatial_id) {
@@ -486,7 +484,7 @@ static void set_flags_and_fb_idx_for_temporal_mode3(VP9_COMP *const cpi) {
cpi->lst_fb_idx = spatial_id - 1;
cpi->gld_fb_idx = spatial_id;
} else {
cpi->gld_fb_idx = spatial_id - 1;
cpi->gld_fb_idx = spatial_id - 1;
}
} else {
cpi->gld_fb_idx = 0;
@@ -515,9 +513,11 @@ static void set_flags_and_fb_idx_for_temporal_mode2(VP9_COMP *const cpi) {
spatial_id = cpi->svc.spatial_layer_id = cpi->svc.spatial_layer_to_encode;
temporal_id = cpi->svc.temporal_layer_id =
cpi->svc.layer_context[cpi->svc.spatial_layer_id *
cpi->svc.number_temporal_layers].current_video_frame_in_layer & 1;
cpi->svc.number_temporal_layers]
.current_video_frame_in_layer &
1;
cpi->ext_refresh_last_frame = cpi->ext_refresh_golden_frame =
cpi->ext_refresh_alt_ref_frame = 0;
cpi->ext_refresh_alt_ref_frame = 0;
if (!temporal_id) {
cpi->ext_refresh_frame_flags_pending = 1;
cpi->ext_refresh_last_frame = 1;
@@ -548,7 +548,7 @@ static void set_flags_and_fb_idx_for_temporal_mode2(VP9_COMP *const cpi) {
cpi->lst_fb_idx = spatial_id - 1;
cpi->gld_fb_idx = spatial_id;
} else {
cpi->gld_fb_idx = spatial_id - 1;
cpi->gld_fb_idx = spatial_id - 1;
}
} else {
cpi->gld_fb_idx = 0;
@@ -567,8 +567,8 @@ static void set_flags_and_fb_idx_for_temporal_mode_noLayering(
VP9_COMP *const cpi) {
int spatial_id;
spatial_id = cpi->svc.spatial_layer_id = cpi->svc.spatial_layer_to_encode;
cpi->ext_refresh_last_frame =
cpi->ext_refresh_golden_frame = cpi->ext_refresh_alt_ref_frame = 0;
cpi->ext_refresh_last_frame = cpi->ext_refresh_golden_frame =
cpi->ext_refresh_alt_ref_frame = 0;
cpi->ext_refresh_frame_flags_pending = 1;
cpi->ext_refresh_last_frame = 1;
if (!spatial_id) {
@@ -586,7 +586,7 @@ static void set_flags_and_fb_idx_for_temporal_mode_noLayering(
cpi->lst_fb_idx = spatial_id - 1;
cpi->gld_fb_idx = spatial_id;
} else {
cpi->gld_fb_idx = spatial_id - 1;
cpi->gld_fb_idx = spatial_id - 1;
}
} else {
cpi->gld_fb_idx = 0;
@@ -596,20 +596,19 @@ static void set_flags_and_fb_idx_for_temporal_mode_noLayering(
int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
int width = 0, height = 0;
LAYER_CONTEXT *lc = NULL;
if (cpi->svc.number_spatial_layers > 1)
cpi->svc.use_base_mv = 1;
if (cpi->svc.number_spatial_layers > 1) cpi->svc.use_base_mv = 1;
cpi->svc.force_zero_mode_spatial_ref = 1;
if (cpi->svc.temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_0212) {
set_flags_and_fb_idx_for_temporal_mode3(cpi);
} else if (cpi->svc.temporal_layering_mode ==
VP9E_TEMPORAL_LAYERING_MODE_NOLAYERING) {
VP9E_TEMPORAL_LAYERING_MODE_NOLAYERING) {
set_flags_and_fb_idx_for_temporal_mode_noLayering(cpi);
} else if (cpi->svc.temporal_layering_mode ==
VP9E_TEMPORAL_LAYERING_MODE_0101) {
VP9E_TEMPORAL_LAYERING_MODE_0101) {
set_flags_and_fb_idx_for_temporal_mode2(cpi);
} else if (cpi->svc.temporal_layering_mode ==
VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
// In the BYPASS/flexible mode, the encoder is relying on the application
// to specify, for each spatial layer, the flags and buffer indices for the
// layering.
@@ -633,7 +632,7 @@ int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
cpi->svc.rc_drop_superframe = 0;
lc = &cpi->svc.layer_context[cpi->svc.spatial_layer_id *
cpi->svc.number_temporal_layers +
cpi->svc.number_temporal_layers +
cpi->svc.temporal_layer_id];
// Setting the worst/best_quality via the encoder control: SET_SVC_PARAMETERS,
@@ -641,12 +640,12 @@ int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) {
if (cpi->svc.temporal_layering_mode != VP9E_TEMPORAL_LAYERING_MODE_BYPASS) {
RATE_CONTROL *const lrc = &lc->rc;
lrc->worst_quality = vp9_quantizer_to_qindex(lc->max_q);
lrc->best_quality = vp9_quantizer_to_qindex(lc->min_q);
lrc->best_quality = vp9_quantizer_to_qindex(lc->min_q);
}
get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height,
lc->scaling_factor_num, lc->scaling_factor_den,
&width, &height);
lc->scaling_factor_num, lc->scaling_factor_den, &width,
&height);
if (vp9_set_size_literal(cpi, width, height) != 0)
return VPX_CODEC_INVALID_PARAM;
@@ -677,8 +676,8 @@ int vp9_svc_start_frame(VP9_COMP *const cpi) {
cpi->lst_fb_idx = cpi->svc.spatial_layer_id;
if (cpi->svc.spatial_layer_id == 0)
cpi->gld_fb_idx = (lc->gold_ref_idx >= 0) ?
lc->gold_ref_idx : cpi->lst_fb_idx;
cpi->gld_fb_idx =
(lc->gold_ref_idx >= 0) ? lc->gold_ref_idx : cpi->lst_fb_idx;
else
cpi->gld_fb_idx = cpi->svc.spatial_layer_id - 1;
@@ -692,8 +691,7 @@ int vp9_svc_start_frame(VP9_COMP *const cpi) {
} else {
if (cpi->oxcf.ss_enable_auto_arf[cpi->svc.spatial_layer_id]) {
cpi->alt_fb_idx = lc->alt_ref_idx;
if (!lc->has_alt_frame)
cpi->ref_frame_flags &= (~VP9_ALT_FLAG);
if (!lc->has_alt_frame) cpi->ref_frame_flags &= (~VP9_ALT_FLAG);
} else {
// Find a proper alt_fb_idx for layers that don't have alt ref frame
if (cpi->svc.spatial_layer_id == 0) {
@@ -714,8 +712,8 @@ int vp9_svc_start_frame(VP9_COMP *const cpi) {
}
get_layer_resolution(cpi->oxcf.width, cpi->oxcf.height,
lc->scaling_factor_num, lc->scaling_factor_den,
&width, &height);
lc->scaling_factor_num, lc->scaling_factor_den, &width,
&height);
// Workaround for multiple frame contexts. In some frames we can't use prev_mi
// since its previous frame could be changed during decoding time. The idea is
@@ -740,11 +738,10 @@ int vp9_svc_start_frame(VP9_COMP *const cpi) {
cpi->common.show_frame = 0;
cpi->ref_frame_flags = 0;
cpi->common.frame_type = INTER_FRAME;
cpi->lst_fb_idx =
cpi->gld_fb_idx = cpi->alt_fb_idx = SMALL_FRAME_FB_IDX;
cpi->lst_fb_idx = cpi->gld_fb_idx = cpi->alt_fb_idx =
SMALL_FRAME_FB_IDX;
if (cpi->svc.encode_intra_empty_frame != 0)
cpi->common.intra_only = 1;
if (cpi->svc.encode_intra_empty_frame != 0) cpi->common.intra_only = 1;
width = SMALL_FRAME_WIDTH;
height = SMALL_FRAME_HEIGHT;
@@ -794,12 +791,9 @@ void vp9_free_svc_cyclic_refresh(VP9_COMP *const cpi) {
for (tl = 0; tl < oxcf->ts_number_layers; ++tl) {
int layer = LAYER_IDS_TO_IDX(sl, tl, oxcf->ts_number_layers);
LAYER_CONTEXT *const lc = &svc->layer_context[layer];
if (lc->map)
vpx_free(lc->map);
if (lc->last_coded_q_map)
vpx_free(lc->last_coded_q_map);
if (lc->consec_zero_mv)
vpx_free(lc->consec_zero_mv);
if (lc->map) vpx_free(lc->map);
if (lc->last_coded_q_map) vpx_free(lc->last_coded_q_map);
if (lc->consec_zero_mv) vpx_free(lc->consec_zero_mv);
}
}
}
@@ -820,9 +814,9 @@ void vp9_svc_reset_key_frame(VP9_COMP *const cpi) {
set_flags_and_fb_idx_for_temporal_mode3(cpi);
} else if (svc->temporal_layering_mode ==
VP9E_TEMPORAL_LAYERING_MODE_NOLAYERING) {
set_flags_and_fb_idx_for_temporal_mode_noLayering(cpi);
set_flags_and_fb_idx_for_temporal_mode_noLayering(cpi);
} else if (svc->temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_0101) {
set_flags_and_fb_idx_for_temporal_mode2(cpi);
set_flags_and_fb_idx_for_temporal_mode2(cpi);
}
vp9_update_temporal_layer_framerate(cpi);
vp9_restore_layer_context(cpi);

8
vp9/encoder/vp9_svc_layercontext.h
View File

@@ -35,7 +35,7 @@ typedef struct {
int is_key_frame;
int frames_from_key_frame;
FRAME_TYPE last_frame_type;
struct lookahead_entry *alt_ref_source;
struct lookahead_entry *alt_ref_source;
int alt_ref_idx;
int gold_ref_idx;
int has_alt_frame;
@@ -60,11 +60,7 @@ typedef struct {
int rc_drop_superframe;
// Workaround for multiple frame contexts
enum {
ENCODED = 0,
ENCODING,
NEED_TO_ENCODE
}encode_empty_frame_state;
enum { ENCODED = 0, ENCODING, NEED_TO_ENCODE } encode_empty_frame_state;
struct lookahead_entry empty_frame;
int encode_intra_empty_frame;

340
vp9/encoder/vp9_temporal_filter.c
View File

@@ -31,18 +31,10 @@
static int fixed_divide[512];
static void temporal_filter_predictors_mb_c(MACROBLOCKD *xd,
uint8_t *y_mb_ptr,
uint8_t *u_mb_ptr,
uint8_t *v_mb_ptr,
int stride,
int uv_block_width,
int uv_block_height,
int mv_row,
int mv_col,
uint8_t *pred,
struct scale_factors *scale,
int x, int y) {
static void temporal_filter_predictors_mb_c(
MACROBLOCKD *xd, uint8_t *y_mb_ptr, uint8_t *u_mb_ptr, uint8_t *v_mb_ptr,
int stride, int uv_block_width, int uv_block_height, int mv_row, int mv_col,
uint8_t *pred, struct scale_factors *scale, int x, int y) {
const int which_mv = 0;
const MV mv = { mv_row, mv_col };
const InterpKernel *const kernel = vp9_filter_kernels[EIGHTTAP_SHARP];
@@ -59,74 +51,46 @@ static void temporal_filter_predictors_mb_c(MACROBLOCKD *xd,
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
vp9_highbd_build_inter_predictor(y_mb_ptr, stride,
&pred[0], 16,
&mv,
scale,
16, 16,
which_mv,
kernel, MV_PRECISION_Q3, x, y, xd->bd);
vp9_highbd_build_inter_predictor(y_mb_ptr, stride, &pred[0], 16, &mv, scale,
16, 16, which_mv, kernel, MV_PRECISION_Q3,
x, y, xd->bd);
vp9_highbd_build_inter_predictor(u_mb_ptr, uv_stride,
&pred[256], uv_block_width,
&mv,
scale,
uv_block_width, uv_block_height,
which_mv,
kernel, mv_precision_uv, x, y, xd->bd);
vp9_highbd_build_inter_predictor(u_mb_ptr, uv_stride, &pred[256],
uv_block_width, &mv, scale, uv_block_width,
uv_block_height, which_mv, kernel,
mv_precision_uv, x, y, xd->bd);
vp9_highbd_build_inter_predictor(v_mb_ptr, uv_stride,
&pred[512], uv_block_width,
&mv,
scale,
uv_block_width, uv_block_height,
which_mv,
kernel, mv_precision_uv, x, y, xd->bd);
vp9_highbd_build_inter_predictor(v_mb_ptr, uv_stride, &pred[512],
uv_block_width, &mv, scale, uv_block_width,
uv_block_height, which_mv, kernel,
mv_precision_uv, x, y, xd->bd);
return;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
(void)xd;
vp9_build_inter_predictor(y_mb_ptr, stride,
&pred[0], 16,
&mv,
scale,
16, 16,
which_mv,
kernel, MV_PRECISION_Q3, x, y);
vp9_build_inter_predictor(y_mb_ptr, stride, &pred[0], 16, &mv, scale, 16, 16,
which_mv, kernel, MV_PRECISION_Q3, x, y);
vp9_build_inter_predictor(u_mb_ptr, uv_stride,
&pred[256], uv_block_width,
&mv,
scale,
uv_block_width, uv_block_height,
which_mv,
kernel, mv_precision_uv, x, y);
vp9_build_inter_predictor(u_mb_ptr, uv_stride, &pred[256], uv_block_width,
&mv, scale, uv_block_width, uv_block_height,
which_mv, kernel, mv_precision_uv, x, y);
vp9_build_inter_predictor(v_mb_ptr, uv_stride,
&pred[512], uv_block_width,
&mv,
scale,
uv_block_width, uv_block_height,
which_mv,
kernel, mv_precision_uv, x, y);
vp9_build_inter_predictor(v_mb_ptr, uv_stride, &pred[512], uv_block_width,
&mv, scale, uv_block_width, uv_block_height,
which_mv, kernel, mv_precision_uv, x, y);
}
void vp9_temporal_filter_init(void) {
int i;
fixed_divide[0] = 0;
for (i = 1; i < 512; ++i)
fixed_divide[i] = 0x80000 / i;
for (i = 1; i < 512; ++i) fixed_divide[i] = 0x80000 / i;
}
void vp9_temporal_filter_apply_c(uint8_t *frame1,
unsigned int stride,
uint8_t *frame2,
unsigned int block_width,
unsigned int block_height,
int strength,
int filter_weight,
unsigned int *accumulator,
void vp9_temporal_filter_apply_c(uint8_t *frame1, unsigned int stride,
uint8_t *frame2, unsigned int block_width,
unsigned int block_height, int strength,
int filter_weight, unsigned int *accumulator,
uint16_t *count) {
unsigned int i, j, k;
int modifier;
@@ -146,10 +110,10 @@ void vp9_temporal_filter_apply_c(uint8_t *frame1,
int row = (int)i + idy;
int col = (int)j + idx;
if (row >= 0 && row < (int)block_height &&
col >= 0 && col < (int)block_width) {
if (row >= 0 && row < (int)block_height && col >= 0 &&
col < (int)block_width) {
int diff = frame1[byte + idy * (int)stride + idx] -
frame2[idy * (int)block_width + idx];
frame2[idy * (int)block_width + idx];
diff_sse[index] = diff * diff;
++index;
}
@@ -159,19 +123,17 @@ void vp9_temporal_filter_apply_c(uint8_t *frame1,
assert(index > 0);
modifier = 0;
for (idx = 0; idx < 9; ++idx)
modifier += diff_sse[idx];
for (idx = 0; idx < 9; ++idx) modifier += diff_sse[idx];
modifier *= 3;
modifier /= index;
++frame2;
modifier += rounding;
modifier += rounding;
modifier >>= strength;
if (modifier > 16)
modifier = 16;
if (modifier > 16) modifier = 16;
modifier = 16 - modifier;
modifier *= filter_weight;
@@ -187,15 +149,10 @@ void vp9_temporal_filter_apply_c(uint8_t *frame1,
}
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
unsigned int stride,
uint8_t *frame2_8,
unsigned int block_width,
unsigned int block_height,
int strength,
int filter_weight,
unsigned int *accumulator,
uint16_t *count) {
void vp9_highbd_temporal_filter_apply_c(
uint8_t *frame1_8, unsigned int stride, uint8_t *frame2_8,
unsigned int block_width, unsigned int block_height, int strength,
int filter_weight, unsigned int *accumulator, uint16_t *count) {
uint16_t *frame1 = CONVERT_TO_SHORTPTR(frame1_8);
uint16_t *frame2 = CONVERT_TO_SHORTPTR(frame2_8);
unsigned int i, j, k;
@@ -214,10 +171,10 @@ void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
int row = (int)i + idy;
int col = (int)j + idx;
if (row >= 0 && row < (int)block_height &&
col >= 0 && col < (int)block_width) {
if (row >= 0 && row < (int)block_height && col >= 0 &&
col < (int)block_width) {
int diff = frame1[byte + idy * (int)stride + idx] -
frame2[idy * (int)block_width + idx];
frame2[idy * (int)block_width + idx];
diff_sse[index] = diff * diff;
++index;
}
@@ -226,8 +183,7 @@ void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
assert(index > 0);
modifier = 0;
for (idx = 0; idx < 9; ++idx)
modifier += diff_sse[idx];
for (idx = 0; idx < 9; ++idx) modifier += diff_sse[idx];
modifier *= 3;
modifier /= index;
@@ -236,8 +192,7 @@ void vp9_highbd_temporal_filter_apply_c(uint8_t *frame1_8,
modifier += rounding;
modifier >>= strength;
if (modifier > 16)
modifier = 16;
if (modifier > 16) modifier = 16;
modifier = 16 - modifier;
modifier *= filter_weight;
@@ -268,7 +223,7 @@ static int temporal_filter_find_matching_mb_c(VP9_COMP *cpi,
uint32_t sse;
int cost_list[5];
MV best_ref_mv1 = {0, 0};
MV best_ref_mv1 = { 0, 0 };
MV best_ref_mv1_full; /* full-pixel value of best_ref_mv1 */
MV *ref_mv = &x->e_mbd.mi[0]->bmi[0].as_mv[0].as_mv;
@@ -295,15 +250,11 @@ static int temporal_filter_find_matching_mb_c(VP9_COMP *cpi,
mv_sf->search_method = old_search_method;
// Ignore mv costing by sending NULL pointer instead of cost array
bestsme = cpi->find_fractional_mv_step(x, ref_mv,
&best_ref_mv1,
cpi->common.allow_high_precision_mv,
x->errorperbit,
&cpi->fn_ptr[BLOCK_16X16],
0, mv_sf->subpel_iters_per_step,
cond_cost_list(cpi, cost_list),
NULL, NULL,
&distortion, &sse, NULL, 0, 0);
bestsme = cpi->find_fractional_mv_step(
x, ref_mv, &best_ref_mv1, cpi->common.allow_high_precision_mv,
x->errorperbit, &cpi->fn_ptr[BLOCK_16X16], 0,
mv_sf->subpel_iters_per_step, cond_cost_list(cpi, cost_list), NULL, NULL,
&distortion, &sse, NULL, 0, 0);
// Restore input state
x->plane[0].src = src;
@@ -314,8 +265,7 @@ static int temporal_filter_find_matching_mb_c(VP9_COMP *cpi,
static void temporal_filter_iterate_c(VP9_COMP *cpi,
YV12_BUFFER_CONFIG **frames,
int frame_count,
int alt_ref_index,
int frame_count, int alt_ref_index,
int strength,
struct scale_factors *scale) {
int byte;
@@ -332,17 +282,17 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
YV12_BUFFER_CONFIG *f = frames[alt_ref_index];
uint8_t *dst1, *dst2;
#if CONFIG_VP9_HIGHBITDEPTH
DECLARE_ALIGNED(16, uint16_t, predictor16[16 * 16 * 3]);
DECLARE_ALIGNED(16, uint8_t, predictor8[16 * 16 * 3]);
DECLARE_ALIGNED(16, uint16_t, predictor16[16 * 16 * 3]);
DECLARE_ALIGNED(16, uint8_t, predictor8[16 * 16 * 3]);
uint8_t *predictor;
#else
DECLARE_ALIGNED(16, uint8_t, predictor[16 * 16 * 3]);
DECLARE_ALIGNED(16, uint8_t, predictor[16 * 16 * 3]);
#endif
const int mb_uv_height = 16 >> mbd->plane[1].subsampling_y;
const int mb_uv_width = 16 >> mbd->plane[1].subsampling_x;
const int mb_uv_width = 16 >> mbd->plane[1].subsampling_x;
// Save input state
uint8_t* input_buffer[MAX_MB_PLANE];
uint8_t *input_buffer[MAX_MB_PLANE];
int i;
#if CONFIG_VP9_HIGHBITDEPTH
if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
@@ -352,8 +302,7 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
}
#endif
for (i = 0; i < MAX_MB_PLANE; i++)
input_buffer[i] = mbd->plane[i].pre[0].buf;
for (i = 0; i < MAX_MB_PLANE; i++) input_buffer[i] = mbd->plane[i].pre[0].buf;
for (mb_row = 0; mb_row < mb_rows; mb_row++) {
// Source frames are extended to 16 pixels. This is different than
@@ -368,8 +317,8 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
// To keep the mv in play for both Y and UV planes the max that it
// can be on a border is therefore 16 - (2*VP9_INTERP_EXTEND+1).
cpi->td.mb.mv_row_min = -((mb_row * 16) + (17 - 2 * VP9_INTERP_EXTEND));
cpi->td.mb.mv_row_max = ((mb_rows - 1 - mb_row) * 16)
+ (17 - 2 * VP9_INTERP_EXTEND);
cpi->td.mb.mv_row_max =
((mb_rows - 1 - mb_row) * 16) + (17 - 2 * VP9_INTERP_EXTEND);
for (mb_col = 0; mb_col < mb_cols; mb_col++) {
int i, j, k;
@@ -379,15 +328,14 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
memset(count, 0, 16 * 16 * 3 * sizeof(count[0]));
cpi->td.mb.mv_col_min = -((mb_col * 16) + (17 - 2 * VP9_INTERP_EXTEND));
cpi->td.mb.mv_col_max = ((mb_cols - 1 - mb_col) * 16)
+ (17 - 2 * VP9_INTERP_EXTEND);
cpi->td.mb.mv_col_max =
((mb_cols - 1 - mb_col) * 16) + (17 - 2 * VP9_INTERP_EXTEND);
for (frame = 0; frame < frame_count; frame++) {
const int thresh_low = 10000;
const int thresh_low = 10000;
const int thresh_high = 20000;
if (frames[frame] == NULL)
continue;
if (frames[frame] == NULL) continue;
mbd->mi[0]->bmi[0].as_mv[0].as_mv.row = 0;
mbd->mi[0]->bmi[0].as_mv[0].as_mv.col = 0;
@@ -396,86 +344,69 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
filter_weight = 2;
} else {
// Find best match in this frame by MC
int err = temporal_filter_find_matching_mb_c(cpi,
frames[alt_ref_index]->y_buffer + mb_y_offset,
frames[frame]->y_buffer + mb_y_offset,
frames[frame]->y_stride);
int err = temporal_filter_find_matching_mb_c(
cpi, frames[alt_ref_index]->y_buffer + mb_y_offset,
frames[frame]->y_buffer + mb_y_offset, frames[frame]->y_stride);
// Assign higher weight to matching MB if its error
// score is lower. If not applying MC default behavior
// is to weight all MBs equal.
filter_weight = err < thresh_low
? 2 : err < thresh_high ? 1 : 0;
filter_weight = err < thresh_low ? 2 : err < thresh_high ? 1 : 0;
}
if (filter_weight != 0) {
// Construct the predictors
temporal_filter_predictors_mb_c(mbd,
frames[frame]->y_buffer + mb_y_offset,
temporal_filter_predictors_mb_c(
mbd, frames[frame]->y_buffer + mb_y_offset,
frames[frame]->u_buffer + mb_uv_offset,
frames[frame]->v_buffer + mb_uv_offset,
frames[frame]->y_stride,
mb_uv_width, mb_uv_height,
mbd->mi[0]->bmi[0].as_mv[0].as_mv.row,
mbd->mi[0]->bmi[0].as_mv[0].as_mv.col,
predictor, scale,
frames[frame]->v_buffer + mb_uv_offset, frames[frame]->y_stride,
mb_uv_width, mb_uv_height, mbd->mi[0]->bmi[0].as_mv[0].as_mv.row,
mbd->mi[0]->bmi[0].as_mv[0].as_mv.col, predictor, scale,
mb_col * 16, mb_row * 16);
#if CONFIG_VP9_HIGHBITDEPTH
if (mbd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
int adj_strength = strength + 2 * (mbd->bd - 8);
// Apply the filter (YUV)
vp9_highbd_temporal_filter_apply_c(f->y_buffer + mb_y_offset,
f->y_stride,
predictor, 16, 16, adj_strength,
filter_weight,
accumulator, count);
vp9_highbd_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
f->uv_stride, predictor + 256,
mb_uv_width, mb_uv_height,
adj_strength, filter_weight,
accumulator + 256, count + 256);
vp9_highbd_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
f->uv_stride, predictor + 512,
mb_uv_width, mb_uv_height,
adj_strength, filter_weight,
accumulator + 512, count + 512);
vp9_highbd_temporal_filter_apply_c(
f->y_buffer + mb_y_offset, f->y_stride, predictor, 16, 16,
adj_strength, filter_weight, accumulator, count);
vp9_highbd_temporal_filter_apply_c(
f->u_buffer + mb_uv_offset, f->uv_stride, predictor + 256,
mb_uv_width, mb_uv_height, adj_strength, filter_weight,
accumulator + 256, count + 256);
vp9_highbd_temporal_filter_apply_c(
f->v_buffer + mb_uv_offset, f->uv_stride, predictor + 512,
mb_uv_width, mb_uv_height, adj_strength, filter_weight,
accumulator + 512, count + 512);
} else {
// Apply the filter (YUV)
vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
predictor, 16, 16,
strength, filter_weight,
accumulator, count);
vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset,
f->uv_stride,
predictor + 256,
mb_uv_width, mb_uv_height, strength,
filter_weight, accumulator + 256,
count + 256);
vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset,
f->uv_stride,
predictor + 512,
mb_uv_width, mb_uv_height, strength,
filter_weight, accumulator + 512,
count + 512);
predictor, 16, 16, strength,
filter_weight, accumulator, count);
vp9_temporal_filter_apply_c(
f->u_buffer + mb_uv_offset, f->uv_stride, predictor + 256,
mb_uv_width, mb_uv_height, strength, filter_weight,
accumulator + 256, count + 256);
vp9_temporal_filter_apply_c(
f->v_buffer + mb_uv_offset, f->uv_stride, predictor + 512,
mb_uv_width, mb_uv_height, strength, filter_weight,
accumulator + 512, count + 512);
}
#else
// Apply the filter (YUV)
// TODO(jingning): Need SIMD optimization for this.
vp9_temporal_filter_apply_c(f->y_buffer + mb_y_offset, f->y_stride,
predictor, 16, 16,
strength, filter_weight,
accumulator, count);
predictor, 16, 16, strength,
filter_weight, accumulator, count);
vp9_temporal_filter_apply_c(f->u_buffer + mb_uv_offset, f->uv_stride,
predictor + 256,
mb_uv_width, mb_uv_height, strength,
filter_weight, accumulator + 256,
count + 256);
predictor + 256, mb_uv_width,
mb_uv_height, strength, filter_weight,
accumulator + 256, count + 256);
vp9_temporal_filter_apply_c(f->v_buffer + mb_uv_offset, f->uv_stride,
predictor + 512,
mb_uv_width, mb_uv_height, strength,
filter_weight, accumulator + 512,
count + 512);
predictor + 512, mb_uv_width,
mb_uv_height, strength, filter_weight,
accumulator + 512, count + 512);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}
@@ -630,13 +561,11 @@ static void temporal_filter_iterate_c(VP9_COMP *cpi,
}
// Restore input state
for (i = 0; i < MAX_MB_PLANE; i++)
mbd->plane[i].pre[0].buf = input_buffer[i];
for (i = 0; i < MAX_MB_PLANE; i++) mbd->plane[i].pre[0].buf = input_buffer[i];
}
// Apply buffer limits and context specific adjustments to arnr filter.
static void adjust_arnr_filter(VP9_COMP *cpi,
int distance, int group_boost,
static void adjust_arnr_filter(VP9_COMP *cpi, int distance, int group_boost,
int *arnr_frames, int *arnr_strength) {
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
const int frames_after_arf =
@@ -647,8 +576,7 @@ static void adjust_arnr_filter(VP9_COMP *cpi,
// Context dependent two pass adjustment to strength.
if (oxcf->pass == 2) {
base_strength =
oxcf->arnr_strength + cpi->twopass.arnr_strength_adjustment;
base_strength = oxcf->arnr_strength + cpi->twopass.arnr_strength_adjustment;
// Clip to allowed range.
base_strength = VPXMIN(6, VPXMAX(0, base_strength));
} else {
@@ -656,34 +584,30 @@ static void adjust_arnr_filter(VP9_COMP *cpi,
}
// Define the forward and backwards filter limits for this arnr group.
if (frames_fwd > frames_after_arf)
frames_fwd = frames_after_arf;
if (frames_fwd > distance)
frames_fwd = distance;
if (frames_fwd > frames_after_arf) frames_fwd = frames_after_arf;
if (frames_fwd > distance) frames_fwd = distance;
frames_bwd = frames_fwd;
// For even length filter there is one more frame backward
// than forward: e.g. len=6 ==> bbbAff, len=7 ==> bbbAfff.
if (frames_bwd < distance)
frames_bwd += (oxcf->arnr_max_frames + 1) & 0x1;
if (frames_bwd < distance) frames_bwd += (oxcf->arnr_max_frames + 1) & 0x1;
// Set the baseline active filter size.
frames = frames_bwd + 1 + frames_fwd;
// Adjust the strength based on active max q.
if (cpi->common.current_video_frame > 1)
q = ((int)vp9_convert_qindex_to_q(
cpi->rc.avg_frame_qindex[INTER_FRAME], cpi->common.bit_depth));
q = ((int)vp9_convert_qindex_to_q(cpi->rc.avg_frame_qindex[INTER_FRAME],
cpi->common.bit_depth));
else
q = ((int)vp9_convert_qindex_to_q(
cpi->rc.avg_frame_qindex[KEY_FRAME], cpi->common.bit_depth));
q = ((int)vp9_convert_qindex_to_q(cpi->rc.avg_frame_qindex[KEY_FRAME],
cpi->common.bit_depth));
if (q > 16) {
strength = base_strength;
} else {
strength = base_strength - ((16 - q) / 2);
if (strength < 0)
strength = 0;
if (strength < 0) strength = 0;
}
// Adjust number of frames in filter and strength based on gf boost level.
@@ -719,7 +643,7 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
int frames_to_blur_backward;
int frames_to_blur_forward;
struct scale_factors sf;
YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS] = {NULL};
YV12_BUFFER_CONFIG *frames[MAX_LAG_BUFFERS] = { NULL };
// Apply context specific adjustments to the arnr filter parameters.
adjust_arnr_filter(cpi, distance, rc->gfu_boost, &frames_to_blur, &strength);
@@ -730,8 +654,8 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
// Setup frame pointers, NULL indicates frame not included in filter.
for (frame = 0; frame < frames_to_blur; ++frame) {
const int which_buffer = start_frame - frame;
struct lookahead_entry *buf = vp9_lookahead_peek(cpi->lookahead,
which_buffer);
struct lookahead_entry *buf =
vp9_lookahead_peek(cpi->lookahead, which_buffer);
frames[frames_to_blur - 1 - frame] = &buf->img;
}
@@ -744,16 +668,13 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
int frame_used = 0;
#if CONFIG_VP9_HIGHBITDEPTH
vp9_setup_scale_factors_for_frame(
&sf,
get_frame_new_buffer(cm)->y_crop_width,
&sf, get_frame_new_buffer(cm)->y_crop_width,
get_frame_new_buffer(cm)->y_crop_height,
get_frame_new_buffer(cm)->y_crop_width,
get_frame_new_buffer(cm)->y_crop_height,
cm->use_highbitdepth);
get_frame_new_buffer(cm)->y_crop_height, cm->use_highbitdepth);
#else
vp9_setup_scale_factors_for_frame(
&sf,
get_frame_new_buffer(cm)->y_crop_width,
&sf, get_frame_new_buffer(cm)->y_crop_width,
get_frame_new_buffer(cm)->y_crop_height,
get_frame_new_buffer(cm)->y_crop_width,
get_frame_new_buffer(cm)->y_crop_height);
@@ -763,14 +684,13 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
if (cm->mi_cols * MI_SIZE != frames[frame]->y_width ||
cm->mi_rows * MI_SIZE != frames[frame]->y_height) {
if (vpx_realloc_frame_buffer(&cpi->svc.scaled_frames[frame_used],
cm->width, cm->height,
cm->subsampling_x, cm->subsampling_y,
cm->width, cm->height, cm->subsampling_x,
cm->subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cm->use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS,
cm->byte_alignment,
NULL, NULL, NULL)) {
cm->byte_alignment, NULL, NULL, NULL)) {
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to reallocate alt_ref_buffer");
}
@@ -783,20 +703,16 @@ void vp9_temporal_filter(VP9_COMP *cpi, int distance) {
xd->mi = cm->mi_grid_visible;
xd->mi[0] = cm->mi;
} else {
// ARF is produced at the native frame size and resized when coded.
// ARF is produced at the native frame size and resized when coded.
#if CONFIG_VP9_HIGHBITDEPTH
vp9_setup_scale_factors_for_frame(&sf,
frames[0]->y_crop_width,
frames[0]->y_crop_height,
frames[0]->y_crop_width,
frames[0]->y_crop_height,
cm->use_highbitdepth);
vp9_setup_scale_factors_for_frame(
&sf, frames[0]->y_crop_width, frames[0]->y_crop_height,
frames[0]->y_crop_width, frames[0]->y_crop_height,
cm->use_highbitdepth);
#else
vp9_setup_scale_factors_for_frame(&sf,
frames[0]->y_crop_width,
frames[0]->y_crop_height,
frames[0]->y_crop_width,
frames[0]->y_crop_height);
vp9_setup_scale_factors_for_frame(
&sf, frames[0]->y_crop_width, frames[0]->y_crop_height,
frames[0]->y_crop_width, frames[0]->y_crop_height);
#endif // CONFIG_VP9_HIGHBITDEPTH
}
}

527
vp9/encoder/vp9_tokenize.c
View File

@@ -24,62 +24,54 @@
#include "vp9/encoder/vp9_tokenize.h"
static const TOKENVALUE dct_cat_lt_10_value_tokens[] = {
{9, 63}, {9, 61}, {9, 59}, {9, 57}, {9, 55}, {9, 53}, {9, 51}, {9, 49},
{9, 47}, {9, 45}, {9, 43}, {9, 41}, {9, 39}, {9, 37}, {9, 35}, {9, 33},
{9, 31}, {9, 29}, {9, 27}, {9, 25}, {9, 23}, {9, 21}, {9, 19}, {9, 17},
{9, 15}, {9, 13}, {9, 11}, {9, 9}, {9, 7}, {9, 5}, {9, 3}, {9, 1},
{8, 31}, {8, 29}, {8, 27}, {8, 25}, {8, 23}, {8, 21},
{8, 19}, {8, 17}, {8, 15}, {8, 13}, {8, 11}, {8, 9},
{8, 7}, {8, 5}, {8, 3}, {8, 1},
{7, 15}, {7, 13}, {7, 11}, {7, 9}, {7, 7}, {7, 5}, {7, 3}, {7, 1},
{6, 7}, {6, 5}, {6, 3}, {6, 1}, {5, 3}, {5, 1},
{4, 1}, {3, 1}, {2, 1}, {1, 1}, {0, 0},
{1, 0}, {2, 0}, {3, 0}, {4, 0},
{5, 0}, {5, 2}, {6, 0}, {6, 2}, {6, 4}, {6, 6},
{7, 0}, {7, 2}, {7, 4}, {7, 6}, {7, 8}, {7, 10}, {7, 12}, {7, 14},
{8, 0}, {8, 2}, {8, 4}, {8, 6}, {8, 8}, {8, 10}, {8, 12},
{8, 14}, {8, 16}, {8, 18}, {8, 20}, {8, 22}, {8, 24},
{8, 26}, {8, 28}, {8, 30}, {9, 0}, {9, 2},
{9, 4}, {9, 6}, {9, 8}, {9, 10}, {9, 12}, {9, 14}, {9, 16},
{9, 18}, {9, 20}, {9, 22}, {9, 24}, {9, 26}, {9, 28},
{9, 30}, {9, 32}, {9, 34}, {9, 36}, {9, 38}, {9, 40},
{9, 42}, {9, 44}, {9, 46}, {9, 48}, {9, 50}, {9, 52},
{9, 54}, {9, 56}, {9, 58}, {9, 60}, {9, 62}
{ 9, 63 }, { 9, 61 }, { 9, 59 }, { 9, 57 }, { 9, 55 }, { 9, 53 }, { 9, 51 },
{ 9, 49 }, { 9, 47 }, { 9, 45 }, { 9, 43 }, { 9, 41 }, { 9, 39 }, { 9, 37 },
{ 9, 35 }, { 9, 33 }, { 9, 31 }, { 9, 29 }, { 9, 27 }, { 9, 25 }, { 9, 23 },
{ 9, 21 }, { 9, 19 }, { 9, 17 }, { 9, 15 }, { 9, 13 }, { 9, 11 }, { 9, 9 },
{ 9, 7 }, { 9, 5 }, { 9, 3 }, { 9, 1 }, { 8, 31 }, { 8, 29 }, { 8, 27 },
{ 8, 25 }, { 8, 23 }, { 8, 21 }, { 8, 19 }, { 8, 17 }, { 8, 15 }, { 8, 13 },
{ 8, 11 }, { 8, 9 }, { 8, 7 }, { 8, 5 }, { 8, 3 }, { 8, 1 }, { 7, 15 },
{ 7, 13 }, { 7, 11 }, { 7, 9 }, { 7, 7 }, { 7, 5 }, { 7, 3 }, { 7, 1 },
{ 6, 7 }, { 6, 5 }, { 6, 3 }, { 6, 1 }, { 5, 3 }, { 5, 1 }, { 4, 1 },
{ 3, 1 }, { 2, 1 }, { 1, 1 }, { 0, 0 }, { 1, 0 }, { 2, 0 }, { 3, 0 },
{ 4, 0 }, { 5, 0 }, { 5, 2 }, { 6, 0 }, { 6, 2 }, { 6, 4 }, { 6, 6 },
{ 7, 0 }, { 7, 2 }, { 7, 4 }, { 7, 6 }, { 7, 8 }, { 7, 10 }, { 7, 12 },
{ 7, 14 }, { 8, 0 }, { 8, 2 }, { 8, 4 }, { 8, 6 }, { 8, 8 }, { 8, 10 },
{ 8, 12 }, { 8, 14 }, { 8, 16 }, { 8, 18 }, { 8, 20 }, { 8, 22 }, { 8, 24 },
{ 8, 26 }, { 8, 28 }, { 8, 30 }, { 9, 0 }, { 9, 2 }, { 9, 4 }, { 9, 6 },
{ 9, 8 }, { 9, 10 }, { 9, 12 }, { 9, 14 }, { 9, 16 }, { 9, 18 }, { 9, 20 },
{ 9, 22 }, { 9, 24 }, { 9, 26 }, { 9, 28 }, { 9, 30 }, { 9, 32 }, { 9, 34 },
{ 9, 36 }, { 9, 38 }, { 9, 40 }, { 9, 42 }, { 9, 44 }, { 9, 46 }, { 9, 48 },
{ 9, 50 }, { 9, 52 }, { 9, 54 }, { 9, 56 }, { 9, 58 }, { 9, 60 }, { 9, 62 }
};
const TOKENVALUE *vp9_dct_cat_lt_10_value_tokens = dct_cat_lt_10_value_tokens +
(sizeof(dct_cat_lt_10_value_tokens) / sizeof(*dct_cat_lt_10_value_tokens))
/ 2;
const TOKENVALUE *vp9_dct_cat_lt_10_value_tokens =
dct_cat_lt_10_value_tokens +
(sizeof(dct_cat_lt_10_value_tokens) / sizeof(*dct_cat_lt_10_value_tokens)) /
2;
// The corresponding costs of the extrabits for the tokens in the above table
// are stored in the table below. The values are obtained from looking up the
// entry for the specified extrabits in the table corresponding to the token
// (as defined in cost element vp9_extra_bits)
// e.g. {9, 63} maps to cat5_cost[63 >> 1], {1, 1} maps to sign_cost[1 >> 1]
static const int dct_cat_lt_10_value_cost[] = {
3773, 3750, 3704, 3681, 3623, 3600, 3554, 3531,
3432, 3409, 3363, 3340, 3282, 3259, 3213, 3190,
3136, 3113, 3067, 3044, 2986, 2963, 2917, 2894,
2795, 2772, 2726, 2703, 2645, 2622, 2576, 2553,
3197, 3116, 3058, 2977, 2881, 2800,
2742, 2661, 2615, 2534, 2476, 2395,
2299, 2218, 2160, 2079,
2566, 2427, 2334, 2195, 2023, 1884, 1791, 1652,
1893, 1696, 1453, 1256, 1229, 864,
512, 512, 512, 512, 0,
512, 512, 512, 512,
864, 1229, 1256, 1453, 1696, 1893,
1652, 1791, 1884, 2023, 2195, 2334, 2427, 2566,
2079, 2160, 2218, 2299, 2395, 2476, 2534, 2615,
2661, 2742, 2800, 2881, 2977, 3058, 3116, 3197,
2553, 2576, 2622, 2645, 2703, 2726, 2772, 2795,
2894, 2917, 2963, 2986, 3044, 3067, 3113, 3136,
3190, 3213, 3259, 3282, 3340, 3363, 3409, 3432,
3531, 3554, 3600, 3623, 3681, 3704, 3750, 3773,
3773, 3750, 3704, 3681, 3623, 3600, 3554, 3531, 3432, 3409, 3363, 3340, 3282,
3259, 3213, 3190, 3136, 3113, 3067, 3044, 2986, 2963, 2917, 2894, 2795, 2772,
2726, 2703, 2645, 2622, 2576, 2553, 3197, 3116, 3058, 2977, 2881, 2800, 2742,
2661, 2615, 2534, 2476, 2395, 2299, 2218, 2160, 2079, 2566, 2427, 2334, 2195,
2023, 1884, 1791, 1652, 1893, 1696, 1453, 1256, 1229, 864, 512, 512, 512,
512, 0, 512, 512, 512, 512, 864, 1229, 1256, 1453, 1696, 1893, 1652,
1791, 1884, 2023, 2195, 2334, 2427, 2566, 2079, 2160, 2218, 2299, 2395, 2476,
2534, 2615, 2661, 2742, 2800, 2881, 2977, 3058, 3116, 3197, 2553, 2576, 2622,
2645, 2703, 2726, 2772, 2795, 2894, 2917, 2963, 2986, 3044, 3067, 3113, 3136,
3190, 3213, 3259, 3282, 3340, 3363, 3409, 3432, 3531, 3554, 3600, 3623, 3681,
3704, 3750, 3773,
};
const int *vp9_dct_cat_lt_10_value_cost = dct_cat_lt_10_value_cost +
(sizeof(dct_cat_lt_10_value_cost) / sizeof(*dct_cat_lt_10_value_cost))
/ 2;
const int *vp9_dct_cat_lt_10_value_cost =
dct_cat_lt_10_value_cost +
(sizeof(dct_cat_lt_10_value_cost) / sizeof(*dct_cat_lt_10_value_cost)) / 2;
// Array indices are identical to previously-existing CONTEXT_NODE indices
/* clang-format off */
const vpx_tree_index vp9_coef_tree[TREE_SIZE(ENTROPY_TOKENS)] = {
-EOB_TOKEN, 2, // 0 = EOB
-ZERO_TOKEN, 4, // 1 = ZERO
@@ -93,226 +85,229 @@ const vpx_tree_index vp9_coef_tree[TREE_SIZE(ENTROPY_TOKENS)] = {
-CATEGORY3_TOKEN, -CATEGORY4_TOKEN, // 9 = CAT_THREE
-CATEGORY5_TOKEN, -CATEGORY6_TOKEN // 10 = CAT_FIVE
};
/* clang-format on */
static const int16_t zero_cost[] = {0};
static const int16_t sign_cost[1] = {512};
static const int16_t cat1_cost[1 << 1] = {864, 1229};
static const int16_t cat2_cost[1 << 2] = {1256, 1453, 1696, 1893};
static const int16_t cat3_cost[1 << 3] = {1652, 1791, 1884, 2023,
2195, 2334, 2427, 2566};
static const int16_t cat4_cost[1 << 4] = {2079, 2160, 2218, 2299, 2395, 2476,
2534, 2615, 2661, 2742, 2800, 2881,
2977, 3058, 3116, 3197};
static const int16_t zero_cost[] = { 0 };
static const int16_t sign_cost[1] = { 512 };
static const int16_t cat1_cost[1 << 1] = { 864, 1229 };
static const int16_t cat2_cost[1 << 2] = { 1256, 1453, 1696, 1893 };
static const int16_t cat3_cost[1 << 3] = { 1652, 1791, 1884, 2023,
2195, 2334, 2427, 2566 };
static const int16_t cat4_cost[1 << 4] = { 2079, 2160, 2218, 2299, 2395, 2476,
2534, 2615, 2661, 2742, 2800, 2881,
2977, 3058, 3116, 3197 };
static const int16_t cat5_cost[1 << 5] = {
2553, 2576, 2622, 2645, 2703, 2726, 2772, 2795, 2894, 2917, 2963,
2986, 3044, 3067, 3113, 3136, 3190, 3213, 3259, 3282, 3340, 3363,
3409, 3432, 3531, 3554, 3600, 3623, 3681, 3704, 3750, 3773};
2553, 2576, 2622, 2645, 2703, 2726, 2772, 2795, 2894, 2917, 2963,
2986, 3044, 3067, 3113, 3136, 3190, 3213, 3259, 3282, 3340, 3363,
3409, 3432, 3531, 3554, 3600, 3623, 3681, 3704, 3750, 3773
};
const int16_t vp9_cat6_low_cost[256] = {
3378, 3390, 3401, 3413, 3435, 3447, 3458, 3470, 3517, 3529, 3540, 3552,
3574, 3586, 3597, 3609, 3671, 3683, 3694, 3706, 3728, 3740, 3751, 3763,
3810, 3822, 3833, 3845, 3867, 3879, 3890, 3902, 3973, 3985, 3996, 4008,
4030, 4042, 4053, 4065, 4112, 4124, 4135, 4147, 4169, 4181, 4192, 4204,
4266, 4278, 4289, 4301, 4323, 4335, 4346, 4358, 4405, 4417, 4428, 4440,
4462, 4474, 4485, 4497, 4253, 4265, 4276, 4288, 4310, 4322, 4333, 4345,
4392, 4404, 4415, 4427, 4449, 4461, 4472, 4484, 4546, 4558, 4569, 4581,
4603, 4615, 4626, 4638, 4685, 4697, 4708, 4720, 4742, 4754, 4765, 4777,
4848, 4860, 4871, 4883, 4905, 4917, 4928, 4940, 4987, 4999, 5010, 5022,
5044, 5056, 5067, 5079, 5141, 5153, 5164, 5176, 5198, 5210, 5221, 5233,
5280, 5292, 5303, 5315, 5337, 5349, 5360, 5372, 4988, 5000, 5011, 5023,
5045, 5057, 5068, 5080, 5127, 5139, 5150, 5162, 5184, 5196, 5207, 5219,
5281, 5293, 5304, 5316, 5338, 5350, 5361, 5373, 5420, 5432, 5443, 5455,
5477, 5489, 5500, 5512, 5583, 5595, 5606, 5618, 5640, 5652, 5663, 5675,
5722, 5734, 5745, 5757, 5779, 5791, 5802, 5814, 5876, 5888, 5899, 5911,
5933, 5945, 5956, 5968, 6015, 6027, 6038, 6050, 6072, 6084, 6095, 6107,
5863, 5875, 5886, 5898, 5920, 5932, 5943, 5955, 6002, 6014, 6025, 6037,
6059, 6071, 6082, 6094, 6156, 6168, 6179, 6191, 6213, 6225, 6236, 6248,
6295, 6307, 6318, 6330, 6352, 6364, 6375, 6387, 6458, 6470, 6481, 6493,
6515, 6527, 6538, 6550, 6597, 6609, 6620, 6632, 6654, 6666, 6677, 6689,
6751, 6763, 6774, 6786, 6808, 6820, 6831, 6843, 6890, 6902, 6913, 6925,
6947, 6959, 6970, 6982};
3378, 3390, 3401, 3413, 3435, 3447, 3458, 3470, 3517, 3529, 3540, 3552, 3574,
3586, 3597, 3609, 3671, 3683, 3694, 3706, 3728, 3740, 3751, 3763, 3810, 3822,
3833, 3845, 3867, 3879, 3890, 3902, 3973, 3985, 3996, 4008, 4030, 4042, 4053,
4065, 4112, 4124, 4135, 4147, 4169, 4181, 4192, 4204, 4266, 4278, 4289, 4301,
4323, 4335, 4346, 4358, 4405, 4417, 4428, 4440, 4462, 4474, 4485, 4497, 4253,
4265, 4276, 4288, 4310, 4322, 4333, 4345, 4392, 4404, 4415, 4427, 4449, 4461,
4472, 4484, 4546, 4558, 4569, 4581, 4603, 4615, 4626, 4638, 4685, 4697, 4708,
4720, 4742, 4754, 4765, 4777, 4848, 4860, 4871, 4883, 4905, 4917, 4928, 4940,
4987, 4999, 5010, 5022, 5044, 5056, 5067, 5079, 5141, 5153, 5164, 5176, 5198,
5210, 5221, 5233, 5280, 5292, 5303, 5315, 5337, 5349, 5360, 5372, 4988, 5000,
5011, 5023, 5045, 5057, 5068, 5080, 5127, 5139, 5150, 5162, 5184, 5196, 5207,
5219, 5281, 5293, 5304, 5316, 5338, 5350, 5361, 5373, 5420, 5432, 5443, 5455,
5477, 5489, 5500, 5512, 5583, 5595, 5606, 5618, 5640, 5652, 5663, 5675, 5722,
5734, 5745, 5757, 5779, 5791, 5802, 5814, 5876, 5888, 5899, 5911, 5933, 5945,
5956, 5968, 6015, 6027, 6038, 6050, 6072, 6084, 6095, 6107, 5863, 5875, 5886,
5898, 5920, 5932, 5943, 5955, 6002, 6014, 6025, 6037, 6059, 6071, 6082, 6094,
6156, 6168, 6179, 6191, 6213, 6225, 6236, 6248, 6295, 6307, 6318, 6330, 6352,
6364, 6375, 6387, 6458, 6470, 6481, 6493, 6515, 6527, 6538, 6550, 6597, 6609,
6620, 6632, 6654, 6666, 6677, 6689, 6751, 6763, 6774, 6786, 6808, 6820, 6831,
6843, 6890, 6902, 6913, 6925, 6947, 6959, 6970, 6982
};
const int vp9_cat6_high_cost[64] = {
88, 2251, 2727, 4890, 3148, 5311, 5787, 7950, 3666, 5829, 6305,
8468, 6726, 8889, 9365, 11528, 3666, 5829, 6305, 8468, 6726, 8889,
9365, 11528, 7244, 9407, 9883, 12046, 10304, 12467, 12943, 15106, 3666,
5829, 6305, 8468, 6726, 8889, 9365, 11528, 7244, 9407, 9883, 12046,
10304, 12467, 12943, 15106, 7244, 9407, 9883, 12046, 10304, 12467, 12943,
15106, 10822, 12985, 13461, 15624, 13882, 16045, 16521, 18684};
88, 2251, 2727, 4890, 3148, 5311, 5787, 7950, 3666, 5829, 6305,
8468, 6726, 8889, 9365, 11528, 3666, 5829, 6305, 8468, 6726, 8889,
9365, 11528, 7244, 9407, 9883, 12046, 10304, 12467, 12943, 15106, 3666,
5829, 6305, 8468, 6726, 8889, 9365, 11528, 7244, 9407, 9883, 12046,
10304, 12467, 12943, 15106, 7244, 9407, 9883, 12046, 10304, 12467, 12943,
15106, 10822, 12985, 13461, 15624, 13882, 16045, 16521, 18684
};
#if CONFIG_VP9_HIGHBITDEPTH
const int vp9_cat6_high10_high_cost[256] = {
94, 2257, 2733, 4896, 3154, 5317, 5793, 7956, 3672, 5835, 6311,
8474, 6732, 8895, 9371, 11534, 3672, 5835, 6311, 8474, 6732, 8895,
9371, 11534, 7250, 9413, 9889, 12052, 10310, 12473, 12949, 15112, 3672,
5835, 6311, 8474, 6732, 8895, 9371, 11534, 7250, 9413, 9889, 12052,
10310, 12473, 12949, 15112, 7250, 9413, 9889, 12052, 10310, 12473, 12949,
15112, 10828, 12991, 13467, 15630, 13888, 16051, 16527, 18690, 4187, 6350,
6826, 8989, 7247, 9410, 9886, 12049, 7765, 9928, 10404, 12567, 10825,
12988, 13464, 15627, 7765, 9928, 10404, 12567, 10825, 12988, 13464, 15627,
11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 7765, 9928, 10404,
12567, 10825, 12988, 13464, 15627, 11343, 13506, 13982, 16145, 14403, 16566,
17042, 19205, 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 14921,
17084, 17560, 19723, 17981, 20144, 20620, 22783, 4187, 6350, 6826, 8989,
7247, 9410, 9886, 12049, 7765, 9928, 10404, 12567, 10825, 12988, 13464,
15627, 7765, 9928, 10404, 12567, 10825, 12988, 13464, 15627, 11343, 13506,
13982, 16145, 14403, 16566, 17042, 19205, 7765, 9928, 10404, 12567, 10825,
12988, 13464, 15627, 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205,
11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 14921, 17084, 17560,
19723, 17981, 20144, 20620, 22783, 8280, 10443, 10919, 13082, 11340, 13503,
13979, 16142, 11858, 14021, 14497, 16660, 14918, 17081, 17557, 19720, 11858,
14021, 14497, 16660, 14918, 17081, 17557, 19720, 15436, 17599, 18075, 20238,
18496, 20659, 21135, 23298, 11858, 14021, 14497, 16660, 14918, 17081, 17557,
19720, 15436, 17599, 18075, 20238, 18496, 20659, 21135, 23298, 15436, 17599,
18075, 20238, 18496, 20659, 21135, 23298, 19014, 21177, 21653, 23816, 22074,
24237, 24713, 26876};
94, 2257, 2733, 4896, 3154, 5317, 5793, 7956, 3672, 5835, 6311,
8474, 6732, 8895, 9371, 11534, 3672, 5835, 6311, 8474, 6732, 8895,
9371, 11534, 7250, 9413, 9889, 12052, 10310, 12473, 12949, 15112, 3672,
5835, 6311, 8474, 6732, 8895, 9371, 11534, 7250, 9413, 9889, 12052,
10310, 12473, 12949, 15112, 7250, 9413, 9889, 12052, 10310, 12473, 12949,
15112, 10828, 12991, 13467, 15630, 13888, 16051, 16527, 18690, 4187, 6350,
6826, 8989, 7247, 9410, 9886, 12049, 7765, 9928, 10404, 12567, 10825,
12988, 13464, 15627, 7765, 9928, 10404, 12567, 10825, 12988, 13464, 15627,
11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 7765, 9928, 10404,
12567, 10825, 12988, 13464, 15627, 11343, 13506, 13982, 16145, 14403, 16566,
17042, 19205, 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 14921,
17084, 17560, 19723, 17981, 20144, 20620, 22783, 4187, 6350, 6826, 8989,
7247, 9410, 9886, 12049, 7765, 9928, 10404, 12567, 10825, 12988, 13464,
15627, 7765, 9928, 10404, 12567, 10825, 12988, 13464, 15627, 11343, 13506,
13982, 16145, 14403, 16566, 17042, 19205, 7765, 9928, 10404, 12567, 10825,
12988, 13464, 15627, 11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205,
11343, 13506, 13982, 16145, 14403, 16566, 17042, 19205, 14921, 17084, 17560,
19723, 17981, 20144, 20620, 22783, 8280, 10443, 10919, 13082, 11340, 13503,
13979, 16142, 11858, 14021, 14497, 16660, 14918, 17081, 17557, 19720, 11858,
14021, 14497, 16660, 14918, 17081, 17557, 19720, 15436, 17599, 18075, 20238,
18496, 20659, 21135, 23298, 11858, 14021, 14497, 16660, 14918, 17081, 17557,
19720, 15436, 17599, 18075, 20238, 18496, 20659, 21135, 23298, 15436, 17599,
18075, 20238, 18496, 20659, 21135, 23298, 19014, 21177, 21653, 23816, 22074,
24237, 24713, 26876
};
const int vp9_cat6_high12_high_cost[1024] = {
100, 2263, 2739, 4902, 3160, 5323, 5799, 7962, 3678, 5841, 6317,
8480, 6738, 8901, 9377, 11540, 3678, 5841, 6317, 8480, 6738, 8901,
9377, 11540, 7256, 9419, 9895, 12058, 10316, 12479, 12955, 15118, 3678,
5841, 6317, 8480, 6738, 8901, 9377, 11540, 7256, 9419, 9895, 12058,
10316, 12479, 12955, 15118, 7256, 9419, 9895, 12058, 10316, 12479, 12955,
15118, 10834, 12997, 13473, 15636, 13894, 16057, 16533, 18696, 4193, 6356,
6832, 8995, 7253, 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831,
12994, 13470, 15633, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633,
11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410,
12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572,
17048, 19211, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927,
17090, 17566, 19729, 17987, 20150, 20626, 22789, 4193, 6356, 6832, 8995,
7253, 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831, 12994, 13470,
15633, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349, 13512,
13988, 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410, 12573, 10831,
12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211,
11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927, 17090, 17566,
19729, 17987, 20150, 20626, 22789, 8286, 10449, 10925, 13088, 11346, 13509,
13985, 16148, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864,
14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244,
18502, 20665, 21141, 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563,
19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605,
18081, 20244, 18502, 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080,
24243, 24719, 26882, 4193, 6356, 6832, 8995, 7253, 9416, 9892, 12055,
7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 7771, 9934, 10410,
12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572,
17048, 19211, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349,
13512, 13988, 16151, 14409, 16572, 17048, 19211, 11349, 13512, 13988, 16151,
14409, 16572, 17048, 19211, 14927, 17090, 17566, 19729, 17987, 20150, 20626,
22789, 8286, 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027,
14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924,
17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304,
11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081,
20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665,
21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882, 8286,
10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027, 14503, 16666,
14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924, 17087, 17563,
19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 11864, 14027,
14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502,
20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304,
19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882, 12379, 14542, 15018,
17181, 15439, 17602, 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180,
21656, 23819, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535,
21698, 22174, 24337, 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759,
19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234,
27397, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276,
25752, 27915, 26173, 28336, 28812, 30975, 4193, 6356, 6832, 8995, 7253,
9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633,
7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988,
16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410, 12573, 10831, 12994,
13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 11349,
13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927, 17090, 17566, 19729,
17987, 20150, 20626, 22789, 8286, 10449, 10925, 13088, 11346, 13509, 13985,
16148, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027,
14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502,
20665, 21141, 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726,
15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081,
20244, 18502, 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243,
24719, 26882, 8286, 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864,
14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666,
14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141,
23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605,
18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502,
20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882,
12379, 14542, 15018, 17181, 15439, 17602, 18078, 20241, 15957, 18120, 18596,
20759, 19017, 21180, 21656, 23819, 15957, 18120, 18596, 20759, 19017, 21180,
21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 15957,
18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337,
22595, 24758, 25234, 27397, 19535, 21698, 22174, 24337, 22595, 24758, 25234,
27397, 23113, 25276, 25752, 27915, 26173, 28336, 28812, 30975, 8286, 10449,
10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027, 14503, 16666, 14924,
17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726,
15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 11864, 14027, 14503,
16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665,
21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 19020,
21183, 21659, 23822, 22080, 24243, 24719, 26882, 12379, 14542, 15018, 17181,
15439, 17602, 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180, 21656,
23819, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698,
22174, 24337, 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759, 19017,
21180, 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397,
19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276, 25752,
27915, 26173, 28336, 28812, 30975, 12379, 14542, 15018, 17181, 15439, 17602,
18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 15957,
18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337,
22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759, 19017, 21180, 21656,
23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 19535, 21698,
22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276, 25752, 27915, 26173,
28336, 28812, 30975, 16472, 18635, 19111, 21274, 19532, 21695, 22171, 24334,
20050, 22213, 22689, 24852, 23110, 25273, 25749, 27912, 20050, 22213, 22689,
24852, 23110, 25273, 25749, 27912, 23628, 25791, 26267, 28430, 26688, 28851,
29327, 31490, 20050, 22213, 22689, 24852, 23110, 25273, 25749, 27912, 23628,
25791, 26267, 28430, 26688, 28851, 29327, 31490, 23628, 25791, 26267, 28430,
26688, 28851, 29327, 31490, 27206, 29369, 29845, 32008, 30266, 32429, 32905,
35068};
100, 2263, 2739, 4902, 3160, 5323, 5799, 7962, 3678, 5841, 6317,
8480, 6738, 8901, 9377, 11540, 3678, 5841, 6317, 8480, 6738, 8901,
9377, 11540, 7256, 9419, 9895, 12058, 10316, 12479, 12955, 15118, 3678,
5841, 6317, 8480, 6738, 8901, 9377, 11540, 7256, 9419, 9895, 12058,
10316, 12479, 12955, 15118, 7256, 9419, 9895, 12058, 10316, 12479, 12955,
15118, 10834, 12997, 13473, 15636, 13894, 16057, 16533, 18696, 4193, 6356,
6832, 8995, 7253, 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831,
12994, 13470, 15633, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633,
11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410,
12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572,
17048, 19211, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927,
17090, 17566, 19729, 17987, 20150, 20626, 22789, 4193, 6356, 6832, 8995,
7253, 9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831, 12994, 13470,
15633, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349, 13512,
13988, 16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410, 12573, 10831,
12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211,
11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927, 17090, 17566,
19729, 17987, 20150, 20626, 22789, 8286, 10449, 10925, 13088, 11346, 13509,
13985, 16148, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864,
14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244,
18502, 20665, 21141, 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563,
19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605,
18081, 20244, 18502, 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080,
24243, 24719, 26882, 4193, 6356, 6832, 8995, 7253, 9416, 9892, 12055,
7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 7771, 9934, 10410,
12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572,
17048, 19211, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349,
13512, 13988, 16151, 14409, 16572, 17048, 19211, 11349, 13512, 13988, 16151,
14409, 16572, 17048, 19211, 14927, 17090, 17566, 19729, 17987, 20150, 20626,
22789, 8286, 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027,
14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924,
17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304,
11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081,
20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665,
21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882, 8286,
10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027, 14503, 16666,
14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924, 17087, 17563,
19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 11864, 14027,
14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502,
20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304,
19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882, 12379, 14542, 15018,
17181, 15439, 17602, 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180,
21656, 23819, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535,
21698, 22174, 24337, 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759,
19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234,
27397, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276,
25752, 27915, 26173, 28336, 28812, 30975, 4193, 6356, 6832, 8995, 7253,
9416, 9892, 12055, 7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633,
7771, 9934, 10410, 12573, 10831, 12994, 13470, 15633, 11349, 13512, 13988,
16151, 14409, 16572, 17048, 19211, 7771, 9934, 10410, 12573, 10831, 12994,
13470, 15633, 11349, 13512, 13988, 16151, 14409, 16572, 17048, 19211, 11349,
13512, 13988, 16151, 14409, 16572, 17048, 19211, 14927, 17090, 17566, 19729,
17987, 20150, 20626, 22789, 8286, 10449, 10925, 13088, 11346, 13509, 13985,
16148, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027,
14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502,
20665, 21141, 23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726,
15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081,
20244, 18502, 20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243,
24719, 26882, 8286, 10449, 10925, 13088, 11346, 13509, 13985, 16148, 11864,
14027, 14503, 16666, 14924, 17087, 17563, 19726, 11864, 14027, 14503, 16666,
14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665, 21141,
23304, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726, 15442, 17605,
18081, 20244, 18502, 20665, 21141, 23304, 15442, 17605, 18081, 20244, 18502,
20665, 21141, 23304, 19020, 21183, 21659, 23822, 22080, 24243, 24719, 26882,
12379, 14542, 15018, 17181, 15439, 17602, 18078, 20241, 15957, 18120, 18596,
20759, 19017, 21180, 21656, 23819, 15957, 18120, 18596, 20759, 19017, 21180,
21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 15957,
18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337,
22595, 24758, 25234, 27397, 19535, 21698, 22174, 24337, 22595, 24758, 25234,
27397, 23113, 25276, 25752, 27915, 26173, 28336, 28812, 30975, 8286, 10449,
10925, 13088, 11346, 13509, 13985, 16148, 11864, 14027, 14503, 16666, 14924,
17087, 17563, 19726, 11864, 14027, 14503, 16666, 14924, 17087, 17563, 19726,
15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 11864, 14027, 14503,
16666, 14924, 17087, 17563, 19726, 15442, 17605, 18081, 20244, 18502, 20665,
21141, 23304, 15442, 17605, 18081, 20244, 18502, 20665, 21141, 23304, 19020,
21183, 21659, 23822, 22080, 24243, 24719, 26882, 12379, 14542, 15018, 17181,
15439, 17602, 18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180, 21656,
23819, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698,
22174, 24337, 22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759, 19017,
21180, 21656, 23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397,
19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276, 25752,
27915, 26173, 28336, 28812, 30975, 12379, 14542, 15018, 17181, 15439, 17602,
18078, 20241, 15957, 18120, 18596, 20759, 19017, 21180, 21656, 23819, 15957,
18120, 18596, 20759, 19017, 21180, 21656, 23819, 19535, 21698, 22174, 24337,
22595, 24758, 25234, 27397, 15957, 18120, 18596, 20759, 19017, 21180, 21656,
23819, 19535, 21698, 22174, 24337, 22595, 24758, 25234, 27397, 19535, 21698,
22174, 24337, 22595, 24758, 25234, 27397, 23113, 25276, 25752, 27915, 26173,
28336, 28812, 30975, 16472, 18635, 19111, 21274, 19532, 21695, 22171, 24334,
20050, 22213, 22689, 24852, 23110, 25273, 25749, 27912, 20050, 22213, 22689,
24852, 23110, 25273, 25749, 27912, 23628, 25791, 26267, 28430, 26688, 28851,
29327, 31490, 20050, 22213, 22689, 24852, 23110, 25273, 25749, 27912, 23628,
25791, 26267, 28430, 26688, 28851, 29327, 31490, 23628, 25791, 26267, 28430,
26688, 28851, 29327, 31490, 27206, 29369, 29845, 32008, 30266, 32429, 32905,
35068
};
#endif
const vp9_extra_bit vp9_extra_bits[ENTROPY_TOKENS] = {
{0, 0, 0, zero_cost}, // ZERO_TOKEN
{0, 0, 1, sign_cost}, // ONE_TOKEN
{0, 0, 2, sign_cost}, // TWO_TOKEN
{0, 0, 3, sign_cost}, // THREE_TOKEN
{0, 0, 4, sign_cost}, // FOUR_TOKEN
{vp9_cat1_prob, 1, CAT1_MIN_VAL, cat1_cost}, // CATEGORY1_TOKEN
{vp9_cat2_prob, 2, CAT2_MIN_VAL, cat2_cost}, // CATEGORY2_TOKEN
{vp9_cat3_prob, 3, CAT3_MIN_VAL, cat3_cost}, // CATEGORY3_TOKEN
{vp9_cat4_prob, 4, CAT4_MIN_VAL, cat4_cost}, // CATEGORY4_TOKEN
{vp9_cat5_prob, 5, CAT5_MIN_VAL, cat5_cost}, // CATEGORY5_TOKEN
{vp9_cat6_prob, 14, CAT6_MIN_VAL, 0}, // CATEGORY6_TOKEN
{0, 0, 0, zero_cost} // EOB_TOKEN
{ 0, 0, 0, zero_cost }, // ZERO_TOKEN
{ 0, 0, 1, sign_cost }, // ONE_TOKEN
{ 0, 0, 2, sign_cost }, // TWO_TOKEN
{ 0, 0, 3, sign_cost }, // THREE_TOKEN
{ 0, 0, 4, sign_cost }, // FOUR_TOKEN
{ vp9_cat1_prob, 1, CAT1_MIN_VAL, cat1_cost }, // CATEGORY1_TOKEN
{ vp9_cat2_prob, 2, CAT2_MIN_VAL, cat2_cost }, // CATEGORY2_TOKEN
{ vp9_cat3_prob, 3, CAT3_MIN_VAL, cat3_cost }, // CATEGORY3_TOKEN
{ vp9_cat4_prob, 4, CAT4_MIN_VAL, cat4_cost }, // CATEGORY4_TOKEN
{ vp9_cat5_prob, 5, CAT5_MIN_VAL, cat5_cost }, // CATEGORY5_TOKEN
{ vp9_cat6_prob, 14, CAT6_MIN_VAL, 0 }, // CATEGORY6_TOKEN
{ 0, 0, 0, zero_cost } // EOB_TOKEN
};
#if CONFIG_VP9_HIGHBITDEPTH
const vp9_extra_bit vp9_extra_bits_high10[ENTROPY_TOKENS] = {
{0, 0, 0, zero_cost}, // ZERO
{0, 0, 1, sign_cost}, // ONE
{0, 0, 2, sign_cost}, // TWO
{0, 0, 3, sign_cost}, // THREE
{0, 0, 4, sign_cost}, // FOUR
{vp9_cat1_prob, 1, CAT1_MIN_VAL, cat1_cost}, // CAT1
{vp9_cat2_prob, 2, CAT2_MIN_VAL, cat2_cost}, // CAT2
{vp9_cat3_prob, 3, CAT3_MIN_VAL, cat3_cost}, // CAT3
{vp9_cat4_prob, 4, CAT4_MIN_VAL, cat4_cost}, // CAT4
{vp9_cat5_prob, 5, CAT5_MIN_VAL, cat5_cost}, // CAT5
{vp9_cat6_prob_high12 + 2, 16, CAT6_MIN_VAL, 0}, // CAT6
{0, 0, 0, zero_cost} // EOB
{ 0, 0, 0, zero_cost }, // ZERO
{ 0, 0, 1, sign_cost }, // ONE
{ 0, 0, 2, sign_cost }, // TWO
{ 0, 0, 3, sign_cost }, // THREE
{ 0, 0, 4, sign_cost }, // FOUR
{ vp9_cat1_prob, 1, CAT1_MIN_VAL, cat1_cost }, // CAT1
{ vp9_cat2_prob, 2, CAT2_MIN_VAL, cat2_cost }, // CAT2
{ vp9_cat3_prob, 3, CAT3_MIN_VAL, cat3_cost }, // CAT3
{ vp9_cat4_prob, 4, CAT4_MIN_VAL, cat4_cost }, // CAT4
{ vp9_cat5_prob, 5, CAT5_MIN_VAL, cat5_cost }, // CAT5
{ vp9_cat6_prob_high12 + 2, 16, CAT6_MIN_VAL, 0 }, // CAT6
{ 0, 0, 0, zero_cost } // EOB
};
const vp9_extra_bit vp9_extra_bits_high12[ENTROPY_TOKENS] = {
{0, 0, 0, zero_cost}, // ZERO
{0, 0, 1, sign_cost}, // ONE
{0, 0, 2, sign_cost}, // TWO
{0, 0, 3, sign_cost}, // THREE
{0, 0, 4, sign_cost}, // FOUR
{vp9_cat1_prob, 1, CAT1_MIN_VAL, cat1_cost}, // CAT1
{vp9_cat2_prob, 2, CAT2_MIN_VAL, cat2_cost}, // CAT2
{vp9_cat3_prob, 3, CAT3_MIN_VAL, cat3_cost}, // CAT3
{vp9_cat4_prob, 4, CAT4_MIN_VAL, cat4_cost}, // CAT4
{vp9_cat5_prob, 5, CAT5_MIN_VAL, cat5_cost}, // CAT5
{vp9_cat6_prob_high12, 18, CAT6_MIN_VAL, 0}, // CAT6
{0, 0, 0, zero_cost} // EOB
{ 0, 0, 0, zero_cost }, // ZERO
{ 0, 0, 1, sign_cost }, // ONE
{ 0, 0, 2, sign_cost }, // TWO
{ 0, 0, 3, sign_cost }, // THREE
{ 0, 0, 4, sign_cost }, // FOUR
{ vp9_cat1_prob, 1, CAT1_MIN_VAL, cat1_cost }, // CAT1
{ vp9_cat2_prob, 2, CAT2_MIN_VAL, cat2_cost }, // CAT2
{ vp9_cat3_prob, 3, CAT3_MIN_VAL, cat3_cost }, // CAT3
{ vp9_cat4_prob, 4, CAT4_MIN_VAL, cat4_cost }, // CAT4
{ vp9_cat5_prob, 5, CAT5_MIN_VAL, cat5_cost }, // CAT5
{ vp9_cat6_prob_high12, 18, CAT6_MIN_VAL, 0 }, // CAT6
{ 0, 0, 0, zero_cost } // EOB
};
#endif
const struct vp9_token vp9_coef_encodings[ENTROPY_TOKENS] = {
{2, 2}, {6, 3}, {28, 5}, {58, 6}, {59, 6}, {60, 6}, {61, 6}, {124, 7},
{125, 7}, {126, 7}, {127, 7}, {0, 1}
{ 2, 2 }, { 6, 3 }, { 28, 5 }, { 58, 6 }, { 59, 6 }, { 60, 6 },
{ 61, 6 }, { 124, 7 }, { 125, 7 }, { 126, 7 }, { 127, 7 }, { 0, 1 }
};
struct tokenize_b_args {
VP9_COMP *cpi;
ThreadData *td;
@@ -320,16 +315,15 @@ struct tokenize_b_args {
};
static void set_entropy_context_b(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct tokenize_b_args* const args = arg;
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *arg) {
struct tokenize_b_args *const args = arg;
ThreadData *const td = args->td;
MACROBLOCK *const x = &td->mb;
MACROBLOCKD *const xd = &x->e_mbd;
struct macroblock_plane *p = &x->plane[plane];
struct macroblockd_plane *pd = &xd->plane[plane];
vp9_set_contexts(xd, pd, plane_bsize, tx_size, p->eobs[block] > 0,
col, row);
vp9_set_contexts(xd, pd, plane_bsize, tx_size, p->eobs[block] > 0, col, row);
}
static INLINE void add_token(TOKENEXTRA **t, const vpx_prob *context_tree,
@@ -344,8 +338,7 @@ static INLINE void add_token(TOKENEXTRA **t, const vpx_prob *context_tree,
static INLINE void add_token_no_extra(TOKENEXTRA **t,
const vpx_prob *context_tree,
int16_t token,
unsigned int *counts) {
int16_t token, unsigned int *counts) {
(*t)->context_tree = context_tree;
(*t)->token = token;
(*t)++;
@@ -353,9 +346,8 @@ static INLINE void add_token_no_extra(TOKENEXTRA **t,
}
static void tokenize_b(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
struct tokenize_b_args* const args = arg;
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *arg) {
struct tokenize_b_args *const args = arg;
VP9_COMP *cpi = args->cpi;
ThreadData *const td = args->td;
MACROBLOCK *const x = &td->mb;
@@ -367,7 +359,7 @@ static void tokenize_b(int plane, int block, int row, int col,
MODE_INFO *mi = xd->mi[0];
int pt; /* near block/prev token context index */
int c;
TOKENEXTRA *t = *tp; /* store tokens starting here */
TOKENEXTRA *t = *tp; /* store tokens starting here */
int eob = p->eobs[block];
const PLANE_TYPE type = get_plane_type(plane);
const tran_low_t *qcoeff = BLOCK_OFFSET(p->qcoeff, block);
@@ -408,8 +400,7 @@ static void tokenize_b(int plane, int block, int row, int col,
vp9_get_token_extra(v, &token, &extra);
add_token(&t, coef_probs[band[c]][pt], token, extra,
counts[band[c]][pt]);
add_token(&t, coef_probs[band[c]][pt], token, extra, counts[band[c]][pt]);
token_cache[scan[c]] = vp9_pt_energy_class[token];
++c;
@@ -432,8 +423,7 @@ struct is_skippable_args {
};
static void is_skippable(int plane, int block, int row, int col,
BLOCK_SIZE plane_bsize, TX_SIZE tx_size,
void *argv) {
BLOCK_SIZE plane_bsize, TX_SIZE tx_size, void *argv) {
struct is_skippable_args *args = argv;
(void)plane;
(void)plane_bsize;
@@ -447,7 +437,7 @@ static void is_skippable(int plane, int block, int row, int col,
// vp9_foreach_transform_block() and simplify is_skippable().
int vp9_is_skippable_in_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
int result = 1;
struct is_skippable_args args = {x->plane[plane].eobs, &result};
struct is_skippable_args args = { x->plane[plane].eobs, &result };
vp9_foreach_transformed_block_in_plane(&x->e_mbd, bsize, plane, is_skippable,
&args);
return result;
@@ -458,36 +448,35 @@ static void has_high_freq_coeff(int plane, int block, int row, int col,
void *argv) {
struct is_skippable_args *args = argv;
int eobs = (tx_size == TX_4X4) ? 3 : 10;
(void) plane;
(void) plane_bsize;
(void) row;
(void) col;
(void)plane;
(void)plane_bsize;
(void)row;
(void)col;
*(args->skippable) |= (args->eobs[block] > eobs);
}
int vp9_has_high_freq_in_plane(MACROBLOCK *x, BLOCK_SIZE bsize, int plane) {
int result = 0;
struct is_skippable_args args = {x->plane[plane].eobs, &result};
struct is_skippable_args args = { x->plane[plane].eobs, &result };
vp9_foreach_transformed_block_in_plane(&x->e_mbd, bsize, plane,
has_high_freq_coeff, &args);
return result;
}
void vp9_tokenize_sb(VP9_COMP *cpi, ThreadData *td, TOKENEXTRA **t,
int dry_run, int seg_skip, BLOCK_SIZE bsize) {
void vp9_tokenize_sb(VP9_COMP *cpi, ThreadData *td, TOKENEXTRA **t, int dry_run,
int seg_skip, BLOCK_SIZE bsize) {
MACROBLOCK *const x = &td->mb;
MACROBLOCKD *const xd = &x->e_mbd;
MODE_INFO *const mi = xd->mi[0];
const int ctx = vp9_get_skip_context(xd);
struct tokenize_b_args arg = {cpi, td, t};
struct tokenize_b_args arg = { cpi, td, t };
if (seg_skip) {
assert(mi->skip);
}
if (mi->skip) {
if (!dry_run && !seg_skip)
++td->counts->skip[ctx][1];
if (!dry_run && !seg_skip) ++td->counts->skip[ctx][1];
reset_skip_context(xd, bsize);
return;
}

24
vp9/encoder/vp9_tokenize.h
View File

@@ -20,15 +20,14 @@
extern "C" {
#endif
#define EOSB_TOKEN 127 // Not signalled, encoder only
#define EOSB_TOKEN 127 // Not signalled, encoder only
#if CONFIG_VP9_HIGHBITDEPTH
typedef int32_t EXTRABIT;
typedef int32_t EXTRABIT;
#else
typedef int16_t EXTRABIT;
typedef int16_t EXTRABIT;
#endif
typedef struct {
int16_t token;
EXTRABIT extra;
@@ -84,19 +83,19 @@ static INLINE int vp9_get_cost(int16_t token, EXTRABIT extrabits,
const int *cat6_high_table) {
if (token != CATEGORY6_TOKEN)
return vp9_extra_bits[token].cost[extrabits >> 1];
return vp9_cat6_low_cost[(extrabits >> 1) & 0xff]
+ cat6_high_table[extrabits >> 9];
return vp9_cat6_low_cost[(extrabits >> 1) & 0xff] +
cat6_high_table[extrabits >> 9];
}
#if CONFIG_VP9_HIGHBITDEPTH
static INLINE const int* vp9_get_high_cost_table(int bit_depth) {
static INLINE const int *vp9_get_high_cost_table(int bit_depth) {
return bit_depth == 8 ? vp9_cat6_high_cost
: (bit_depth == 10 ? vp9_cat6_high10_high_cost :
vp9_cat6_high12_high_cost);
: (bit_depth == 10 ? vp9_cat6_high10_high_cost
: vp9_cat6_high12_high_cost);
}
#else
static INLINE const int* vp9_get_high_cost_table(int bit_depth) {
(void) bit_depth;
static INLINE const int *vp9_get_high_cost_table(int bit_depth) {
(void)bit_depth;
return vp9_cat6_high_cost;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
@@ -114,8 +113,7 @@ static INLINE void vp9_get_token_extra(int v, int16_t *token, EXTRABIT *extra) {
*extra = vp9_dct_cat_lt_10_value_tokens[v].extra;
}
static INLINE int16_t vp9_get_token(int v) {
if (v >= CAT6_MIN_VAL || v <= -CAT6_MIN_VAL)
return 10;
if (v >= CAT6_MIN_VAL || v <= -CAT6_MIN_VAL) return 10;
return vp9_dct_cat_lt_10_value_tokens[v].token;
}

6
vp9/encoder/vp9_treewriter.h
View File

@@ -18,15 +18,15 @@ extern "C" {
#endif
void vp9_tree_probs_from_distribution(vpx_tree tree,
unsigned int branch_ct[ /* n - 1 */ ][2],
const unsigned int num_events[ /* n */ ]);
unsigned int branch_ct[/* n - 1 */][2],
const unsigned int num_events[/* n */]);
struct vp9_token {
int value;
int len;
};
void vp9_tokens_from_tree(struct vp9_token*, const vpx_tree_index *);
void vp9_tokens_from_tree(struct vp9_token *, const vpx_tree_index *);
static INLINE void vp9_write_tree(vpx_writer *w, const vpx_tree_index *tree,
const vpx_prob *probs, int bits, int len,

153
vp9/encoder/x86/vp9_dct_intrin_sse2.c
View File

@@ -78,8 +78,8 @@ static void fdct4_sse2(__m128i *in) {
const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
__m128i u[4], v[4];
u[0]=_mm_unpacklo_epi16(in[0], in[1]);
u[1]=_mm_unpacklo_epi16(in[3], in[2]);
u[0] = _mm_unpacklo_epi16(in[0], in[1]);
u[1] = _mm_unpacklo_epi16(in[3], in[2]);
v[0] = _mm_add_epi16(u[0], u[1]);
v[1] = _mm_sub_epi16(u[0], u[1]);
@@ -151,14 +151,12 @@ static void fadst4_sse2(__m128i *in) {
transpose_4x4(in);
}
void vp9_fht4x4_sse2(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_fht4x4_sse2(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
__m128i in[4];
switch (tx_type) {
case DCT_DCT:
vpx_fdct4x4_sse2(input, output, stride);
break;
case DCT_DCT: vpx_fdct4x4_sse2(input, output, stride); break;
case ADST_DCT:
load_buffer_4x4(input, in, stride);
fadst4_sse2(in);
@@ -177,21 +175,18 @@ void vp9_fht4x4_sse2(const int16_t *input, tran_low_t *output,
fadst4_sse2(in);
write_buffer_4x4(output, in);
break;
default:
assert(0);
break;
default: assert(0); break;
}
}
void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
int16_t* coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t* zbin_ptr,
const int16_t* round_ptr, const int16_t* quant_ptr,
const int16_t* quant_shift_ptr, int16_t* qcoeff_ptr,
int16_t* dqcoeff_ptr, const int16_t* dequant_ptr,
uint16_t* eob_ptr,
const int16_t* scan_ptr,
const int16_t* iscan_ptr) {
int16_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr,
int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan_ptr,
const int16_t *iscan_ptr) {
__m128i zero;
int pass;
// Constants
@@ -208,14 +203,14 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
const __m128i k__cospi_m20_p12 = pair_set_epi16(-cospi_20_64, cospi_12_64);
const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
// Load input
__m128i in0 = _mm_load_si128((const __m128i *)(input + 0 * stride));
__m128i in1 = _mm_load_si128((const __m128i *)(input + 1 * stride));
__m128i in2 = _mm_load_si128((const __m128i *)(input + 2 * stride));
__m128i in3 = _mm_load_si128((const __m128i *)(input + 3 * stride));
__m128i in4 = _mm_load_si128((const __m128i *)(input + 4 * stride));
__m128i in5 = _mm_load_si128((const __m128i *)(input + 5 * stride));
__m128i in6 = _mm_load_si128((const __m128i *)(input + 6 * stride));
__m128i in7 = _mm_load_si128((const __m128i *)(input + 7 * stride));
__m128i in0 = _mm_load_si128((const __m128i *)(input + 0 * stride));
__m128i in1 = _mm_load_si128((const __m128i *)(input + 1 * stride));
__m128i in2 = _mm_load_si128((const __m128i *)(input + 2 * stride));
__m128i in3 = _mm_load_si128((const __m128i *)(input + 3 * stride));
__m128i in4 = _mm_load_si128((const __m128i *)(input + 4 * stride));
__m128i in5 = _mm_load_si128((const __m128i *)(input + 5 * stride));
__m128i in6 = _mm_load_si128((const __m128i *)(input + 6 * stride));
__m128i in7 = _mm_load_si128((const __m128i *)(input + 7 * stride));
__m128i *in[8];
int index = 0;
@@ -469,9 +464,9 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
// Setup global values
{
round = _mm_load_si128((const __m128i*)round_ptr);
quant = _mm_load_si128((const __m128i*)quant_ptr);
dequant = _mm_load_si128((const __m128i*)dequant_ptr);
round = _mm_load_si128((const __m128i *)round_ptr);
quant = _mm_load_si128((const __m128i *)quant_ptr);
dequant = _mm_load_si128((const __m128i *)dequant_ptr);
}
{
@@ -503,15 +498,15 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
}
{
@@ -524,8 +519,8 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
iscan0 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs) + 1);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, nzero_coeff0);
iscan1 = _mm_sub_epi16(iscan1, nzero_coeff1);
@@ -568,14 +563,14 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
}
{
@@ -588,8 +583,8 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
iscan0 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs) + 1);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, nzero_coeff0);
iscan1 = _mm_sub_epi16(iscan1, nzero_coeff1);
@@ -615,10 +610,10 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
}
} else {
do {
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
n_coeffs += 8 * 2;
} while (n_coeffs < 0);
*eob_ptr = 0;
@@ -628,14 +623,14 @@ void vp9_fdct8x8_quant_sse2(const int16_t *input, int stride,
// load 8x8 array
static INLINE void load_buffer_8x8(const int16_t *input, __m128i *in,
int stride) {
in[0] = _mm_load_si128((const __m128i *)(input + 0 * stride));
in[1] = _mm_load_si128((const __m128i *)(input + 1 * stride));
in[2] = _mm_load_si128((const __m128i *)(input + 2 * stride));
in[3] = _mm_load_si128((const __m128i *)(input + 3 * stride));
in[4] = _mm_load_si128((const __m128i *)(input + 4 * stride));
in[5] = _mm_load_si128((const __m128i *)(input + 5 * stride));
in[6] = _mm_load_si128((const __m128i *)(input + 6 * stride));
in[7] = _mm_load_si128((const __m128i *)(input + 7 * stride));
in[0] = _mm_load_si128((const __m128i *)(input + 0 * stride));
in[1] = _mm_load_si128((const __m128i *)(input + 1 * stride));
in[2] = _mm_load_si128((const __m128i *)(input + 2 * stride));
in[3] = _mm_load_si128((const __m128i *)(input + 3 * stride));
in[4] = _mm_load_si128((const __m128i *)(input + 4 * stride));
in[5] = _mm_load_si128((const __m128i *)(input + 5 * stride));
in[6] = _mm_load_si128((const __m128i *)(input + 6 * stride));
in[7] = _mm_load_si128((const __m128i *)(input + 7 * stride));
in[0] = _mm_slli_epi16(in[0], 2);
in[1] = _mm_slli_epi16(in[1], 2);
@@ -930,14 +925,14 @@ static void fadst8_sse2(__m128i *in) {
__m128i in0, in1, in2, in3, in4, in5, in6, in7;
// properly aligned for butterfly input
in0 = in[7];
in1 = in[0];
in2 = in[5];
in3 = in[2];
in4 = in[3];
in5 = in[4];
in6 = in[1];
in7 = in[6];
in0 = in[7];
in1 = in[0];
in2 = in[5];
in3 = in[2];
in4 = in[3];
in5 = in[4];
in6 = in[1];
in7 = in[6];
// column transformation
// stage 1
@@ -1135,14 +1130,12 @@ static void fadst8_sse2(__m128i *in) {
array_transpose_8x8(in, in);
}
void vp9_fht8x8_sse2(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_fht8x8_sse2(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
__m128i in[8];
switch (tx_type) {
case DCT_DCT:
vpx_fdct8x8_sse2(input, output, stride);
break;
case DCT_DCT: vpx_fdct8x8_sse2(input, output, stride); break;
case ADST_DCT:
load_buffer_8x8(input, in, stride);
fadst8_sse2(in);
@@ -1164,13 +1157,11 @@ void vp9_fht8x8_sse2(const int16_t *input, tran_low_t *output,
right_shift_8x8(in, 1);
write_buffer_8x8(output, in, 8);
break;
default:
assert(0);
break;
default: assert(0); break;
}
}
static INLINE void load_buffer_16x16(const int16_t* input, __m128i *in0,
static INLINE void load_buffer_16x16(const int16_t *input, __m128i *in0,
__m128i *in1, int stride) {
// load first 8 columns
load_buffer_8x8(input, in0, stride);
@@ -1530,13 +1521,13 @@ static void fdct16_8col(__m128i *in) {
v[14] = _mm_srai_epi32(u[14], DCT_CONST_BITS);
v[15] = _mm_srai_epi32(u[15], DCT_CONST_BITS);
in[1] = _mm_packs_epi32(v[0], v[1]);
in[9] = _mm_packs_epi32(v[2], v[3]);
in[5] = _mm_packs_epi32(v[4], v[5]);
in[1] = _mm_packs_epi32(v[0], v[1]);
in[9] = _mm_packs_epi32(v[2], v[3]);
in[5] = _mm_packs_epi32(v[4], v[5]);
in[13] = _mm_packs_epi32(v[6], v[7]);
in[3] = _mm_packs_epi32(v[8], v[9]);
in[3] = _mm_packs_epi32(v[8], v[9]);
in[11] = _mm_packs_epi32(v[10], v[11]);
in[7] = _mm_packs_epi32(v[12], v[13]);
in[7] = _mm_packs_epi32(v[12], v[13]);
in[15] = _mm_packs_epi32(v[14], v[15]);
}
@@ -2022,14 +2013,12 @@ static void fadst16_sse2(__m128i *in0, __m128i *in1) {
array_transpose_16x16(in0, in1);
}
void vp9_fht16x16_sse2(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
void vp9_fht16x16_sse2(const int16_t *input, tran_low_t *output, int stride,
int tx_type) {
__m128i in0[16], in1[16];
switch (tx_type) {
case DCT_DCT:
vpx_fdct16x16_sse2(input, output, stride);
break;
case DCT_DCT: vpx_fdct16x16_sse2(input, output, stride); break;
case ADST_DCT:
load_buffer_16x16(input, in0, in1, stride);
fadst16_sse2(in0, in1);
@@ -2051,8 +2040,6 @@ void vp9_fht16x16_sse2(const int16_t *input, tran_low_t *output,
fadst16_sse2(in0, in1);
write_buffer_16x16(output, in0, in1, 16);
break;
default:
assert(0);
break;
default: assert(0); break;
}
}

80
vp9/encoder/x86/vp9_dct_ssse3.c
View File

@@ -15,16 +15,12 @@
#include "vpx_dsp/x86/inv_txfm_sse2.h"
#include "vpx_dsp/x86/txfm_common_sse2.h"
void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
int16_t* coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t* zbin_ptr,
const int16_t* round_ptr, const int16_t* quant_ptr,
const int16_t* quant_shift_ptr,
int16_t* qcoeff_ptr,
int16_t* dqcoeff_ptr, const int16_t* dequant_ptr,
uint16_t* eob_ptr,
const int16_t* scan_ptr,
const int16_t* iscan_ptr) {
void vp9_fdct8x8_quant_ssse3(
const int16_t *input, int stride, int16_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr,
const int16_t *quant_ptr, const int16_t *quant_shift_ptr,
int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan_ptr, const int16_t *iscan_ptr) {
__m128i zero;
int pass;
// Constants
@@ -42,14 +38,14 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
const __m128i k__cospi_m20_p12 = pair_set_epi16(-cospi_20_64, cospi_12_64);
const __m128i k__DCT_CONST_ROUNDING = _mm_set1_epi32(DCT_CONST_ROUNDING);
// Load input
__m128i in0 = _mm_load_si128((const __m128i *)(input + 0 * stride));
__m128i in1 = _mm_load_si128((const __m128i *)(input + 1 * stride));
__m128i in2 = _mm_load_si128((const __m128i *)(input + 2 * stride));
__m128i in3 = _mm_load_si128((const __m128i *)(input + 3 * stride));
__m128i in4 = _mm_load_si128((const __m128i *)(input + 4 * stride));
__m128i in5 = _mm_load_si128((const __m128i *)(input + 5 * stride));
__m128i in6 = _mm_load_si128((const __m128i *)(input + 6 * stride));
__m128i in7 = _mm_load_si128((const __m128i *)(input + 7 * stride));
__m128i in0 = _mm_load_si128((const __m128i *)(input + 0 * stride));
__m128i in1 = _mm_load_si128((const __m128i *)(input + 1 * stride));
__m128i in2 = _mm_load_si128((const __m128i *)(input + 2 * stride));
__m128i in3 = _mm_load_si128((const __m128i *)(input + 3 * stride));
__m128i in4 = _mm_load_si128((const __m128i *)(input + 4 * stride));
__m128i in5 = _mm_load_si128((const __m128i *)(input + 5 * stride));
__m128i in6 = _mm_load_si128((const __m128i *)(input + 6 * stride));
__m128i in7 = _mm_load_si128((const __m128i *)(input + 7 * stride));
__m128i *in[8];
int index = 0;
@@ -298,9 +294,9 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
// Setup global values
{
round = _mm_load_si128((const __m128i*)round_ptr);
quant = _mm_load_si128((const __m128i*)quant_ptr);
dequant = _mm_load_si128((const __m128i*)dequant_ptr);
round = _mm_load_si128((const __m128i *)round_ptr);
quant = _mm_load_si128((const __m128i *)quant_ptr);
dequant = _mm_load_si128((const __m128i *)dequant_ptr);
}
{
@@ -332,15 +328,15 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
}
{
@@ -353,8 +349,8 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
iscan0 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs) + 1);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, nzero_coeff0);
iscan1 = _mm_sub_epi16(iscan1, nzero_coeff1);
@@ -388,7 +384,7 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
nzflag = _mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff0, thr)) |
_mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff1, thr));
_mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff1, thr));
if (nzflag) {
qcoeff0 = _mm_adds_epi16(qcoeff0, round);
@@ -402,20 +398,20 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
} else {
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
}
}
@@ -429,8 +425,8 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
iscan0 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs) + 1);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, nzero_coeff0);
iscan1 = _mm_sub_epi16(iscan1, nzero_coeff1);
@@ -456,10 +452,10 @@ void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,
}
} else {
do {
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
n_coeffs += 8 * 2;
} while (n_coeffs < 0);
*eob_ptr = 0;

119
vp9/encoder/x86/vp9_denoiser_sse2.c
View File

@@ -37,17 +37,11 @@ static INLINE int sum_diff_16x1(__m128i acc_diff) {
}
// Denoise a 16x1 vector.
static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
const uint8_t *mc_running_avg_y,
uint8_t *running_avg_y,
const __m128i *k_0,
const __m128i *k_4,
const __m128i *k_8,
const __m128i *k_16,
const __m128i *l3,
const __m128i *l32,
const __m128i *l21,
__m128i acc_diff) {
static INLINE __m128i vp9_denoiser_16x1_sse2(
const uint8_t *sig, const uint8_t *mc_running_avg_y, uint8_t *running_avg_y,
const __m128i *k_0, const __m128i *k_4, const __m128i *k_8,
const __m128i *k_16, const __m128i *l3, const __m128i *l32,
const __m128i *l21, __m128i acc_diff) {
// Calculate differences
const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0]));
const __m128i v_mc_running_avg_y =
@@ -69,7 +63,7 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
__m128i adj2 = _mm_and_si128(mask2, *l32);
const __m128i adj1 = _mm_and_si128(mask1, *l21);
const __m128i adj0 = _mm_and_si128(mask0, clamped_absdiff);
__m128i adj, padj, nadj;
__m128i adj, padj, nadj;
// Combine the adjustments and get absolute adjustments.
adj2 = _mm_add_epi8(adj2, adj1);
@@ -95,9 +89,8 @@ static INLINE __m128i vp9_denoiser_16x1_sse2(const uint8_t *sig,
// Denoise a 16x1 vector with a weaker filter.
static INLINE __m128i vp9_denoiser_adj_16x1_sse2(
const uint8_t *sig, const uint8_t *mc_running_avg_y,
uint8_t *running_avg_y, const __m128i k_0,
const __m128i k_delta, __m128i acc_diff) {
const uint8_t *sig, const uint8_t *mc_running_avg_y, uint8_t *running_avg_y,
const __m128i k_0, const __m128i k_delta, __m128i acc_diff) {
__m128i v_running_avg_y = _mm_loadu_si128((__m128i *)(&running_avg_y[0]));
// Calculate differences.
const __m128i v_sig = _mm_loadu_si128((const __m128i *)(&sig[0]));
@@ -108,8 +101,7 @@ static INLINE __m128i vp9_denoiser_adj_16x1_sse2(
// Obtain the sign. FF if diff is negative.
const __m128i diff_sign = _mm_cmpeq_epi8(pdiff, k_0);
// Clamp absolute difference to delta to get the adjustment.
const __m128i adj =
_mm_min_epu8(_mm_or_si128(pdiff, ndiff), k_delta);
const __m128i adj = _mm_min_epu8(_mm_or_si128(pdiff, ndiff), k_delta);
// Restore the sign and get positive and negative adjustments.
__m128i padj, nadj;
padj = _mm_andnot_si128(diff_sign, adj);
@@ -126,14 +118,17 @@ static INLINE __m128i vp9_denoiser_adj_16x1_sse2(
}
// Denoise 8x8 and 8x16 blocks.
static int vp9_denoiser_NxM_sse2_small(
const uint8_t *sig, int sig_stride, const uint8_t *mc_running_avg_y,
int mc_avg_y_stride, uint8_t *running_avg_y, int avg_y_stride,
int increase_denoising, BLOCK_SIZE bs, int motion_magnitude, int width) {
static int vp9_denoiser_NxM_sse2_small(const uint8_t *sig, int sig_stride,
const uint8_t *mc_running_avg_y,
int mc_avg_y_stride,
uint8_t *running_avg_y, int avg_y_stride,
int increase_denoising, BLOCK_SIZE bs,
int motion_magnitude, int width) {
int sum_diff_thresh, r, sum_diff = 0;
const int shift_inc = (increase_denoising &&
motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
1 : 0;
const int shift_inc =
(increase_denoising && motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD)
? 1
: 0;
uint8_t sig_buffer[8][16], mc_running_buffer[8][16], running_buffer[8][16];
__m128i acc_diff = _mm_setzero_si128();
const __m128i k_0 = _mm_setzero_si128();
@@ -153,15 +148,13 @@ static int vp9_denoiser_NxM_sse2_small(
memcpy(sig_buffer[r], sig, width);
memcpy(sig_buffer[r] + width, sig + sig_stride, width);
memcpy(mc_running_buffer[r], mc_running_avg_y, width);
memcpy(mc_running_buffer[r] + width,
mc_running_avg_y + mc_avg_y_stride, width);
memcpy(mc_running_buffer[r] + width, mc_running_avg_y + mc_avg_y_stride,
width);
memcpy(running_buffer[r], running_avg_y, width);
memcpy(running_buffer[r] + width, running_avg_y + avg_y_stride, width);
acc_diff = vp9_denoiser_16x1_sse2(sig_buffer[r],
mc_running_buffer[r],
running_buffer[r],
&k_0, &k_4, &k_8, &k_16,
&l3, &l32, &l21, acc_diff);
acc_diff = vp9_denoiser_16x1_sse2(sig_buffer[r], mc_running_buffer[r],
running_buffer[r], &k_0, &k_4, &k_8,
&k_16, &l3, &l32, &l21, acc_diff);
memcpy(running_avg_y, running_buffer[r], width);
memcpy(running_avg_y + avg_y_stride, running_buffer[r] + width, width);
// Update pointers for next iteration.
@@ -184,19 +177,19 @@ static int vp9_denoiser_NxM_sse2_small(
// The delta is set by the excess of absolute pixel diff over the
// threshold.
const int delta = ((abs(sum_diff) - sum_diff_thresh) >>
num_pels_log2_lookup[bs]) + 1;
const int delta =
((abs(sum_diff) - sum_diff_thresh) >> num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
const __m128i k_delta = _mm_set1_epi8(delta);
running_avg_y -= avg_y_stride * (b_height << 1);
for (r = 0; r < b_height; ++r) {
acc_diff = vp9_denoiser_adj_16x1_sse2(
sig_buffer[r], mc_running_buffer[r], running_buffer[r],
k_0, k_delta, acc_diff);
sig_buffer[r], mc_running_buffer[r], running_buffer[r], k_0,
k_delta, acc_diff);
memcpy(running_avg_y, running_buffer[r], width);
memcpy(running_avg_y + avg_y_stride,
running_buffer[r] + width, width);
memcpy(running_avg_y + avg_y_stride, running_buffer[r] + width,
width);
// Update pointers for next iteration.
running_avg_y += (avg_y_stride << 1);
}
@@ -216,14 +209,14 @@ static int vp9_denoiser_NxM_sse2_small(
static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
const uint8_t *mc_running_avg_y,
int mc_avg_y_stride,
uint8_t *running_avg_y,
int avg_y_stride,
uint8_t *running_avg_y, int avg_y_stride,
int increase_denoising, BLOCK_SIZE bs,
int motion_magnitude) {
int sum_diff_thresh, r, c, sum_diff = 0;
const int shift_inc = (increase_denoising &&
motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD) ?
1 : 0;
const int shift_inc =
(increase_denoising && motion_magnitude <= MOTION_MAGNITUDE_THRESHOLD)
? 1
: 0;
__m128i acc_diff[4][4];
const __m128i k_0 = _mm_setzero_si128();
const __m128i k_4 = _mm_set1_epi8(4 + shift_inc);
@@ -248,9 +241,9 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
for (r = 0; r < b_height; ++r) {
for (c = 0; c < b_width_shift4; ++c) {
acc_diff[c][r>>4] = vp9_denoiser_16x1_sse2(
sig, mc_running_avg_y, running_avg_y, &k_0, &k_4,
&k_8, &k_16, &l3, &l32, &l21, acc_diff[c][r>>4]);
acc_diff[c][r >> 4] = vp9_denoiser_16x1_sse2(
sig, mc_running_avg_y, running_avg_y, &k_0, &k_4, &k_8, &k_16, &l3,
&l32, &l21, acc_diff[c][r >> 4]);
// Update pointers for next iteration.
sig += 16;
mc_running_avg_y += 16;
@@ -259,7 +252,7 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
if ((r & 0xf) == 0xf || (bs == BLOCK_16X8 && r == 7)) {
for (c = 0; c < b_width_shift4; ++c) {
sum_diff += sum_diff_16x1(acc_diff[c][r>>4]);
sum_diff += sum_diff_16x1(acc_diff[c][r >> 4]);
}
}
@@ -272,8 +265,8 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
{
sum_diff_thresh = total_adj_strong_thresh(bs, increase_denoising);
if (abs(sum_diff) > sum_diff_thresh) {
const int delta = ((abs(sum_diff) - sum_diff_thresh) >>
num_pels_log2_lookup[bs]) + 1;
const int delta =
((abs(sum_diff) - sum_diff_thresh) >> num_pels_log2_lookup[bs]) + 1;
// Only apply the adjustment for max delta up to 3.
if (delta < 4) {
@@ -284,9 +277,9 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
sum_diff = 0;
for (r = 0; r < b_height; ++r) {
for (c = 0; c < b_width_shift4; ++c) {
acc_diff[c][r>>4] = vp9_denoiser_adj_16x1_sse2(
sig, mc_running_avg_y, running_avg_y, k_0,
k_delta, acc_diff[c][r>>4]);
acc_diff[c][r >> 4] =
vp9_denoiser_adj_16x1_sse2(sig, mc_running_avg_y, running_avg_y,
k_0, k_delta, acc_diff[c][r >> 4]);
// Update pointers for next iteration.
sig += 16;
mc_running_avg_y += 16;
@@ -295,7 +288,7 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
if ((r & 0xf) == 0xf || (bs == BLOCK_16X8 && r == 7)) {
for (c = 0; c < b_width_shift4; ++c) {
sum_diff += sum_diff_16x1(acc_diff[c][r>>4]);
sum_diff += sum_diff_16x1(acc_diff[c][r >> 4]);
}
}
sig = sig - b_width + sig_stride;
@@ -314,27 +307,21 @@ static int vp9_denoiser_NxM_sse2_big(const uint8_t *sig, int sig_stride,
}
int vp9_denoiser_filter_sse2(const uint8_t *sig, int sig_stride,
const uint8_t *mc_avg,
int mc_avg_stride,
const uint8_t *mc_avg, int mc_avg_stride,
uint8_t *avg, int avg_stride,
int increase_denoising,
BLOCK_SIZE bs,
int increase_denoising, BLOCK_SIZE bs,
int motion_magnitude) {
// Rank by frequency of the block type to have an early termination.
if (bs == BLOCK_16X16 || bs == BLOCK_32X32 || bs == BLOCK_64X64 ||
bs == BLOCK_16X32 || bs == BLOCK_16X8 || bs == BLOCK_32X16 ||
bs == BLOCK_32X64 || bs == BLOCK_64X32) {
return vp9_denoiser_NxM_sse2_big(sig, sig_stride,
mc_avg, mc_avg_stride,
avg, avg_stride,
increase_denoising,
bs, motion_magnitude);
return vp9_denoiser_NxM_sse2_big(sig, sig_stride, mc_avg, mc_avg_stride,
avg, avg_stride, increase_denoising, bs,
motion_magnitude);
} else if (bs == BLOCK_8X8 || bs == BLOCK_8X16) {
return vp9_denoiser_NxM_sse2_small(sig, sig_stride,
mc_avg, mc_avg_stride,
avg, avg_stride,
increase_denoising,
bs, motion_magnitude, 8);
return vp9_denoiser_NxM_sse2_small(sig, sig_stride, mc_avg, mc_avg_stride,
avg, avg_stride, increase_denoising, bs,
motion_magnitude, 8);
} else {
return COPY_BLOCK;
}

66
vp9/encoder/x86/vp9_diamond_search_sad_avx.c
View File

@@ -9,7 +9,7 @@
*/
#if defined(_MSC_VER)
# include <intrin.h>
#include <intrin.h>
#endif
#include <emmintrin.h>
#include <smmintrin.h>
@@ -19,11 +19,11 @@
#include "vpx_ports/mem.h"
#ifdef __GNUC__
# define LIKELY(v) __builtin_expect(v, 1)
# define UNLIKELY(v) __builtin_expect(v, 0)
#define LIKELY(v) __builtin_expect(v, 1)
#define UNLIKELY(v) __builtin_expect(v, 0)
#else
# define LIKELY(v) (v)
# define UNLIKELY(v) (v)
#define LIKELY(v) (v)
#define UNLIKELY(v) (v)
#endif
static INLINE int_mv pack_int_mv(int16_t row, int16_t col) {
@@ -40,19 +40,19 @@ static INLINE MV_JOINT_TYPE get_mv_joint(const int_mv mv) {
return mv.as_int == 0 ? 0 : 1;
}
static INLINE int mv_cost(const int_mv mv,
const int *joint_cost, int *const comp_cost[2]) {
return joint_cost[get_mv_joint(mv)] +
comp_cost[0][mv.as_mv.row] + comp_cost[1][mv.as_mv.col];
static INLINE int mv_cost(const int_mv mv, const int *joint_cost,
int *const comp_cost[2]) {
return joint_cost[get_mv_joint(mv)] + comp_cost[0][mv.as_mv.row] +
comp_cost[1][mv.as_mv.col];
}
static int mvsad_err_cost(const MACROBLOCK *x, const int_mv mv, const MV *ref,
int sad_per_bit) {
const int_mv diff = pack_int_mv(mv.as_mv.row - ref->row,
mv.as_mv.col - ref->col);
return ROUND_POWER_OF_TWO((unsigned)mv_cost(diff, x->nmvjointsadcost,
x->nmvsadcost) *
sad_per_bit, VP9_PROB_COST_SHIFT);
const int_mv diff =
pack_int_mv(mv.as_mv.row - ref->row, mv.as_mv.col - ref->col);
return ROUND_POWER_OF_TWO(
(unsigned)mv_cost(diff, x->nmvjointsadcost, x->nmvsadcost) * sad_per_bit,
VP9_PROB_COST_SHIFT);
}
/*****************************************************************************
@@ -71,10 +71,9 @@ static int mvsad_err_cost(const MACROBLOCK *x, const int_mv mv, const MV *ref,
* which does not rely on these properties. *
*****************************************************************************/
int vp9_diamond_search_sad_avx(const MACROBLOCK *x,
const search_site_config *cfg,
MV *ref_mv, MV *best_mv, int search_param,
int sad_per_bit, int *num00,
const vp9_variance_fn_ptr_t *fn_ptr,
const search_site_config *cfg, MV *ref_mv,
MV *best_mv, int search_param, int sad_per_bit,
int *num00, const vp9_variance_fn_ptr_t *fn_ptr,
const MV *center_mv) {
const int_mv maxmv = pack_int_mv(x->mv_row_max, x->mv_col_max);
const __m128i v_max_mv_w = _mm_set1_epi32(maxmv.as_int);
@@ -91,12 +90,12 @@ int vp9_diamond_search_sad_avx(const MACROBLOCK *x,
// 0 = initial step (MAX_FIRST_STEP) pel
// 1 = (MAX_FIRST_STEP/2) pel,
// 2 = (MAX_FIRST_STEP/4) pel...
const MV *ss_mv = &cfg->ss_mv[cfg->searches_per_step * search_param];
const MV *ss_mv = &cfg->ss_mv[cfg->searches_per_step * search_param];
const intptr_t *ss_os = &cfg->ss_os[cfg->searches_per_step * search_param];
const int tot_steps = cfg->total_steps - search_param;
const int_mv fcenter_mv = pack_int_mv(center_mv->row >> 3,
center_mv->col >> 3);
const int_mv fcenter_mv =
pack_int_mv(center_mv->row >> 3, center_mv->col >> 3);
const __m128i vfcmv = _mm_set1_epi32(fcenter_mv.as_int);
const int ref_row = clamp(ref_mv->row, minmv.as_mv.row, maxmv.as_mv.row);
@@ -109,8 +108,8 @@ int vp9_diamond_search_sad_avx(const MACROBLOCK *x,
const int what_stride = x->plane[0].src.stride;
const int in_what_stride = x->e_mbd.plane[0].pre[0].stride;
const uint8_t *const what = x->plane[0].src.buf;
const uint8_t *const in_what = x->e_mbd.plane[0].pre[0].buf +
ref_row * in_what_stride + ref_col;
const uint8_t *const in_what =
x->e_mbd.plane[0].pre[0].buf + ref_row * in_what_stride + ref_col;
// Work out the start point for the search
const uint8_t *best_address = in_what;
@@ -181,10 +180,9 @@ int vp9_diamond_search_sad_avx(const MACROBLOCK *x,
__m128i v_bo10_q = _mm_loadu_si128((const __m128i *)&ss_os[i + 0]);
__m128i v_bo32_q = _mm_loadu_si128((const __m128i *)&ss_os[i + 2]);
// Set the ones falling outside to zero
v_bo10_q = _mm_and_si128(v_bo10_q,
_mm_cvtepi32_epi64(v_inside_d));
v_bo32_q = _mm_and_si128(v_bo32_q,
_mm_unpackhi_epi32(v_inside_d, v_inside_d));
v_bo10_q = _mm_and_si128(v_bo10_q, _mm_cvtepi32_epi64(v_inside_d));
v_bo32_q =
_mm_and_si128(v_bo32_q, _mm_unpackhi_epi32(v_inside_d, v_inside_d));
// Compute the candidate addresses
v_blocka[0] = _mm_add_epi64(v_ba_q, v_bo10_q);
v_blocka[1] = _mm_add_epi64(v_ba_q, v_bo32_q);
@@ -195,9 +193,8 @@ int vp9_diamond_search_sad_avx(const MACROBLOCK *x,
#endif
}
fn_ptr->sdx4df(what, what_stride,
(const uint8_t **)&v_blocka[0], in_what_stride,
(uint32_t*)&v_sad_d);
fn_ptr->sdx4df(what, what_stride, (const uint8_t **)&v_blocka[0],
in_what_stride, (uint32_t *)&v_sad_d);
// Look up the component cost of the residual motion vector
{
@@ -226,11 +223,10 @@ int vp9_diamond_search_sad_avx(const MACROBLOCK *x,
// Now add in the joint cost
{
const __m128i v_sel_d = _mm_cmpeq_epi32(v_diff_mv_w,
_mm_setzero_si128());
const __m128i v_joint_cost_d = _mm_blendv_epi8(v_joint_cost_1_d,
v_joint_cost_0_d,
v_sel_d);
const __m128i v_sel_d =
_mm_cmpeq_epi32(v_diff_mv_w, _mm_setzero_si128());
const __m128i v_joint_cost_d =
_mm_blendv_epi8(v_joint_cost_1_d, v_joint_cost_0_d, v_sel_d);
v_cost_d = _mm_add_epi32(v_cost_d, v_joint_cost_d);
}

12
vp9/encoder/x86/vp9_error_intrin_avx2.c
View File

@@ -13,10 +13,8 @@
#include "./vp9_rtcd.h"
#include "vpx/vpx_integer.h"
int64_t vp9_block_error_avx2(const int16_t *coeff,
const int16_t *dqcoeff,
intptr_t block_size,
int64_t *ssz) {
int64_t vp9_block_error_avx2(const int16_t *coeff, const int16_t *dqcoeff,
intptr_t block_size, int64_t *ssz) {
__m256i sse_reg, ssz_reg, coeff_reg, dqcoeff_reg;
__m256i exp_dqcoeff_lo, exp_dqcoeff_hi, exp_coeff_lo, exp_coeff_hi;
__m256i sse_reg_64hi, ssz_reg_64hi;
@@ -29,7 +27,7 @@ int64_t vp9_block_error_avx2(const int16_t *coeff,
sse_reg = _mm256_set1_epi16(0);
ssz_reg = _mm256_set1_epi16(0);
for (i = 0 ; i < block_size ; i+= 16) {
for (i = 0; i < block_size; i += 16) {
// load 32 bytes from coeff and dqcoeff
coeff_reg = _mm256_loadu_si256((const __m256i *)(coeff + i));
dqcoeff_reg = _mm256_loadu_si256((const __m256i *)(dqcoeff + i));
@@ -66,8 +64,8 @@ int64_t vp9_block_error_avx2(const int16_t *coeff,
_mm256_extractf128_si256(ssz_reg, 1));
// store the results
_mm_storel_epi64((__m128i*)(&sse), sse_reg128);
_mm_storel_epi64((__m128i *)(&sse), sse_reg128);
_mm_storel_epi64((__m128i*)(ssz), ssz_reg128);
_mm_storel_epi64((__m128i *)(ssz), ssz_reg128);
return sse;
}

46
vp9/encoder/x86/vp9_frame_scale_ssse3.c
View File

@@ -19,23 +19,22 @@ extern void vp9_scale_and_extend_frame_c(const YV12_BUFFER_CONFIG *src,
YV12_BUFFER_CONFIG *dst);
static void downsample_2_to_1_ssse3(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
int w, int h) {
uint8_t *dst, ptrdiff_t dst_stride, int w,
int h) {
const __m128i mask = _mm_set1_epi16(0x00FF);
const int max_width = w & ~15;
int y;
for (y = 0; y < h; ++y) {
int x;
for (x = 0; x < max_width; x += 16) {
const __m128i a = _mm_loadu_si128((const __m128i *)(src + x * 2 + 0));
const __m128i a = _mm_loadu_si128((const __m128i *)(src + x * 2 + 0));
const __m128i b = _mm_loadu_si128((const __m128i *)(src + x * 2 + 16));
const __m128i a_and = _mm_and_si128(a, mask);
const __m128i b_and = _mm_and_si128(b, mask);
const __m128i c = _mm_packus_epi16(a_and, b_and);
_mm_storeu_si128((__m128i *)(dst + x), c);
}
for (; x < w; ++x)
dst[x] = src[x * 2];
for (; x < w; ++x) dst[x] = src[x * 2];
src += src_stride * 2;
dst += dst_stride;
}
@@ -47,9 +46,8 @@ static INLINE __m128i filter(const __m128i *const a, const __m128i *const b,
const __m128i *const g, const __m128i *const h) {
const __m128i coeffs_ab =
_mm_set_epi8(6, -1, 6, -1, 6, -1, 6, -1, 6, -1, 6, -1, 6, -1, 6, -1);
const __m128i coeffs_cd =
_mm_set_epi8(78, -19, 78, -19, 78, -19, 78, -19, 78, -19, 78, -19,
78, -19, 78, -19);
const __m128i coeffs_cd = _mm_set_epi8(78, -19, 78, -19, 78, -19, 78, -19, 78,
-19, 78, -19, 78, -19, 78, -19);
const __m128i const64_x16 = _mm_set1_epi16(64);
const __m128i ab = _mm_unpacklo_epi8(*a, *b);
const __m128i cd = _mm_unpacklo_epi8(*c, *d);
@@ -88,8 +86,8 @@ static void eight_tap_row_ssse3(const uint8_t *src, uint8_t *dst, int w) {
}
static void upsample_1_to_2_ssse3(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
int dst_w, int dst_h) {
uint8_t *dst, ptrdiff_t dst_stride, int dst_w,
int dst_h) {
dst_w /= 2;
dst_h /= 2;
{
@@ -116,7 +114,7 @@ static void upsample_1_to_2_ssse3(const uint8_t *src, ptrdiff_t src_stride,
int x;
eight_tap_row_ssse3(src + src_stride * 4 - 3, tmp7, dst_w);
for (x = 0; x < max_width; x += 8) {
const __m128i A = _mm_loadl_epi64((const __m128i *)(src + x));
const __m128i A = _mm_loadl_epi64((const __m128i *)(src + x));
const __m128i B = _mm_loadl_epi64((const __m128i *)(tmp3 + x));
const __m128i AB = _mm_unpacklo_epi8(A, B);
__m128i C, D, CD;
@@ -179,23 +177,23 @@ void vp9_scale_and_extend_frame_ssse3(const YV12_BUFFER_CONFIG *src,
const int dst_uv_h = dst_h / 2;
if (dst_w * 2 == src_w && dst_h * 2 == src_h) {
downsample_2_to_1_ssse3(src->y_buffer, src->y_stride,
dst->y_buffer, dst->y_stride, dst_w, dst_h);
downsample_2_to_1_ssse3(src->u_buffer, src->uv_stride,
dst->u_buffer, dst->uv_stride, dst_uv_w, dst_uv_h);
downsample_2_to_1_ssse3(src->v_buffer, src->uv_stride,
dst->v_buffer, dst->uv_stride, dst_uv_w, dst_uv_h);
downsample_2_to_1_ssse3(src->y_buffer, src->y_stride, dst->y_buffer,
dst->y_stride, dst_w, dst_h);
downsample_2_to_1_ssse3(src->u_buffer, src->uv_stride, dst->u_buffer,
dst->uv_stride, dst_uv_w, dst_uv_h);
downsample_2_to_1_ssse3(src->v_buffer, src->uv_stride, dst->v_buffer,
dst->uv_stride, dst_uv_w, dst_uv_h);
vpx_extend_frame_borders(dst);
} else if (dst_w == src_w * 2 && dst_h == src_h * 2) {
// The upsample() supports widths up to 1920 * 2. If greater, fall back
// to vp9_scale_and_extend_frame_c().
if (dst_w/2 <= 1920) {
upsample_1_to_2_ssse3(src->y_buffer, src->y_stride,
dst->y_buffer, dst->y_stride, dst_w, dst_h);
upsample_1_to_2_ssse3(src->u_buffer, src->uv_stride,
dst->u_buffer, dst->uv_stride, dst_uv_w, dst_uv_h);
upsample_1_to_2_ssse3(src->v_buffer, src->uv_stride,
dst->v_buffer, dst->uv_stride, dst_uv_w, dst_uv_h);
if (dst_w / 2 <= 1920) {
upsample_1_to_2_ssse3(src->y_buffer, src->y_stride, dst->y_buffer,
dst->y_stride, dst_w, dst_h);
upsample_1_to_2_ssse3(src->u_buffer, src->uv_stride, dst->u_buffer,
dst->uv_stride, dst_uv_w, dst_uv_h);
upsample_1_to_2_ssse3(src->v_buffer, src->uv_stride, dst->v_buffer,
dst->uv_stride, dst_uv_w, dst_uv_h);
vpx_extend_frame_borders(dst);
} else {
vp9_scale_and_extend_frame_c(src, dst);

30
vp9/encoder/x86/vp9_highbd_block_error_intrin_sse2.c
View File

@@ -23,41 +23,41 @@ int64_t vp9_highbd_block_error_sse2(tran_low_t *coeff, tran_low_t *dqcoeff,
const int shift = 2 * (bps - 8);
const int rounding = shift > 0 ? 1 << (shift - 1) : 0;
for (i = 0; i < block_size; i+=8) {
for (i = 0; i < block_size; i += 8) {
// Load the data into xmm registers
__m128i mm_coeff = _mm_load_si128((__m128i*) (coeff + i));
__m128i mm_coeff2 = _mm_load_si128((__m128i*) (coeff + i + 4));
__m128i mm_dqcoeff = _mm_load_si128((__m128i*) (dqcoeff + i));
__m128i mm_dqcoeff2 = _mm_load_si128((__m128i*) (dqcoeff + i + 4));
__m128i mm_coeff = _mm_load_si128((__m128i *)(coeff + i));
__m128i mm_coeff2 = _mm_load_si128((__m128i *)(coeff + i + 4));
__m128i mm_dqcoeff = _mm_load_si128((__m128i *)(dqcoeff + i));
__m128i mm_dqcoeff2 = _mm_load_si128((__m128i *)(dqcoeff + i + 4));
// Check if any values require more than 15 bit
max = _mm_set1_epi32(0x3fff);
min = _mm_set1_epi32(0xffffc000);
cmp0 = _mm_xor_si128(_mm_cmpgt_epi32(mm_coeff, max),
_mm_cmplt_epi32(mm_coeff, min));
_mm_cmplt_epi32(mm_coeff, min));
cmp1 = _mm_xor_si128(_mm_cmpgt_epi32(mm_coeff2, max),
_mm_cmplt_epi32(mm_coeff2, min));
_mm_cmplt_epi32(mm_coeff2, min));
cmp2 = _mm_xor_si128(_mm_cmpgt_epi32(mm_dqcoeff, max),
_mm_cmplt_epi32(mm_dqcoeff, min));
_mm_cmplt_epi32(mm_dqcoeff, min));
cmp3 = _mm_xor_si128(_mm_cmpgt_epi32(mm_dqcoeff2, max),
_mm_cmplt_epi32(mm_dqcoeff2, min));
test = _mm_movemask_epi8(_mm_or_si128(_mm_or_si128(cmp0, cmp1),
_mm_or_si128(cmp2, cmp3)));
_mm_cmplt_epi32(mm_dqcoeff2, min));
test = _mm_movemask_epi8(
_mm_or_si128(_mm_or_si128(cmp0, cmp1), _mm_or_si128(cmp2, cmp3)));
if (!test) {
__m128i mm_diff, error_sse2, sqcoeff_sse2;;
__m128i mm_diff, error_sse2, sqcoeff_sse2;
mm_coeff = _mm_packs_epi32(mm_coeff, mm_coeff2);
mm_dqcoeff = _mm_packs_epi32(mm_dqcoeff, mm_dqcoeff2);
mm_diff = _mm_sub_epi16(mm_coeff, mm_dqcoeff);
error_sse2 = _mm_madd_epi16(mm_diff, mm_diff);
sqcoeff_sse2 = _mm_madd_epi16(mm_coeff, mm_coeff);
_mm_storeu_si128((__m128i*)temp, error_sse2);
_mm_storeu_si128((__m128i *)temp, error_sse2);
error = error + temp[0] + temp[1] + temp[2] + temp[3];
_mm_storeu_si128((__m128i*)temp, sqcoeff_sse2);
_mm_storeu_si128((__m128i *)temp, sqcoeff_sse2);
sqcoeff += temp[0] + temp[1] + temp[2] + temp[3];
} else {
for (j = 0; j < 8; j++) {
const int64_t diff = coeff[i + j] - dqcoeff[i + j];
error += diff * diff;
error += diff * diff;
sqcoeff += (int64_t)coeff[i + j] * (int64_t)coeff[i + j];
}
}

71
vp9/encoder/x86/vp9_quantize_sse2.c
View File

@@ -14,14 +14,13 @@
#include "./vp9_rtcd.h"
#include "vpx/vpx_integer.h"
void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t* zbin_ptr,
const int16_t* round_ptr, const int16_t* quant_ptr,
const int16_t* quant_shift_ptr, int16_t* qcoeff_ptr,
int16_t* dqcoeff_ptr, const int16_t* dequant_ptr,
uint16_t* eob_ptr,
const int16_t* scan_ptr,
const int16_t* iscan_ptr) {
void vp9_quantize_fp_sse2(const int16_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, int16_t *qcoeff_ptr,
int16_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan_ptr,
const int16_t *iscan_ptr) {
__m128i zero;
__m128i thr;
int16_t nzflag;
@@ -44,9 +43,9 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
// Setup global values
{
round = _mm_load_si128((const __m128i*)round_ptr);
quant = _mm_load_si128((const __m128i*)quant_ptr);
dequant = _mm_load_si128((const __m128i*)dequant_ptr);
round = _mm_load_si128((const __m128i *)round_ptr);
quant = _mm_load_si128((const __m128i *)quant_ptr);
dequant = _mm_load_si128((const __m128i *)dequant_ptr);
}
{
@@ -54,8 +53,8 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
__m128i qcoeff0, qcoeff1;
__m128i qtmp0, qtmp1;
// Do DC and first 15 AC
coeff0 = _mm_load_si128((const __m128i*)(coeff_ptr + n_coeffs));
coeff1 = _mm_load_si128((const __m128i*)(coeff_ptr + n_coeffs) + 1);
coeff0 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs));
coeff1 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs) + 1);
// Poor man's sign extract
coeff0_sign = _mm_srai_epi16(coeff0, 15);
@@ -78,15 +77,15 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
dequant = _mm_unpackhi_epi64(dequant, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
}
{
@@ -99,8 +98,8 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
iscan0 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs) + 1);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, nzero_coeff0);
iscan1 = _mm_sub_epi16(iscan1, nzero_coeff1);
@@ -121,8 +120,8 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
__m128i qcoeff0, qcoeff1;
__m128i qtmp0, qtmp1;
coeff0 = _mm_load_si128((const __m128i*)(coeff_ptr + n_coeffs));
coeff1 = _mm_load_si128((const __m128i*)(coeff_ptr + n_coeffs) + 1);
coeff0 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs));
coeff1 = _mm_load_si128((const __m128i *)(coeff_ptr + n_coeffs) + 1);
// Poor man's sign extract
coeff0_sign = _mm_srai_epi16(coeff0, 15);
@@ -133,7 +132,7 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
nzflag = _mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff0, thr)) |
_mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff1, thr));
_mm_movemask_epi8(_mm_cmpgt_epi16(qcoeff1, thr));
if (nzflag) {
qcoeff0 = _mm_adds_epi16(qcoeff0, round);
@@ -147,20 +146,20 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
qcoeff0 = _mm_sub_epi16(qcoeff0, coeff0_sign);
qcoeff1 = _mm_sub_epi16(qcoeff1, coeff1_sign);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), qcoeff0);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, qcoeff1);
coeff0 = _mm_mullo_epi16(qcoeff0, dequant);
coeff1 = _mm_mullo_epi16(qcoeff1, dequant);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), coeff0);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, coeff1);
} else {
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
}
}
@@ -174,8 +173,8 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
zero_coeff1 = _mm_cmpeq_epi16(coeff1, zero);
nzero_coeff0 = _mm_cmpeq_epi16(zero_coeff0, zero);
nzero_coeff1 = _mm_cmpeq_epi16(zero_coeff1, zero);
iscan0 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i*)(iscan_ptr + n_coeffs) + 1);
iscan0 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs));
iscan1 = _mm_load_si128((const __m128i *)(iscan_ptr + n_coeffs) + 1);
// Add one to convert from indices to counts
iscan0 = _mm_sub_epi16(iscan0, nzero_coeff0);
iscan1 = _mm_sub_epi16(iscan1, nzero_coeff1);
@@ -200,10 +199,10 @@ void vp9_quantize_fp_sse2(const int16_t* coeff_ptr, intptr_t n_coeffs,
}
} else {
do {
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i*)(qcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(dqcoeff_ptr + n_coeffs) + 1, zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs), zero);
_mm_store_si128((__m128i *)(qcoeff_ptr + n_coeffs) + 1, zero);
n_coeffs += 8 * 2;
} while (n_coeffs < 0);
*eob_ptr = 0;